[clang] 992a64a - [Clang][SVE] Change LLVM representation of ACLE tuple types to be struct based. (#108008)

[via cfe-commits]

Author: Paul Walker
Date: 2024-09-13T12:45:21+01:00
New Revision: 992a64aad388b4e809ac3123ea1d48fed4f35244

URL: https://github.com/llvm/llvm-project/commit/992a64aad388b4e809ac3123ea1d48fed4f35244
DIFF: https://github.com/llvm/llvm-project/commit/992a64aad388b4e809ac3123ea1d48fed4f35244.diff

LOG: [Clang][SVE] Change LLVM representation of ACLE tuple types to be struct based. (#108008)

This implements our original design now that LLVM is comfortable with
structs and arrays of scalable vector types. All SVE ACLE intrinsics
already use struct types so the effect of this change is purely the
types used for alloca and function parameters.
    
There should be no C/C++ user visible change with this patch.

Added: 
    

Modified: 
    clang/include/clang/Basic/AArch64SVEACLETypes.def
    clang/lib/CodeGen/CGBuiltin.cpp
    clang/lib/CodeGen/CodeGenFunction.h
    clang/lib/CodeGen/CodeGenTypes.cpp
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_clamp.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtl.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtn.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fmlas16.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fp_dots.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_frint.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_int_dots.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x2.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x4.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x2.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x4.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mla.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlal.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlall.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mls.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlsl.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_read.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sqdmulh.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sub.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx2.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx4.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vdot.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_add.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_qrshr.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_rshl.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx2.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx4.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx2.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx4.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx2.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx4.c
    clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_write.c
    clang/test/CodeGen/aarch64-sme2p1-intrinsics/acle_sme2p1_movaz.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_reinterpret-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_reinterpret.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_set2-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_set2.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_set3-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_set3.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_set4-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_set4.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st2.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st3-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st3.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st4-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_st4.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_undef2-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_undef2.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_undef3-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_undef3.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_undef4-bfloat.c
    clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_undef4.c
    clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_luti.c
    clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_tbl2-bfloat.c
    clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_tbl2.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_create2_bool.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_create4_bool.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_get2_bool.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_get4_bool.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_ld1.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_ldnt1.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_loads.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_pext.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_qcvtn.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_qrshr.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_set2_bool.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_set4_bool.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_st1.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_stnt1.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_store.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_undef_bool.c
    clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_while_x2.c
    clang/test/CodeGen/aarch64_neon_sve_bridge_intrinsics/acle_neon_sve_bridge_get_neonq.c
    clang/test/CodeGen/svboolx2_t.cpp
    clang/test/CodeGen/svboolx4_t.cpp
    clang/test/CodeGenCXX/aarch64-mangle-sve-vectors.cpp
    clang/test/CodeGenCXX/aarch64-sve-vector-init.cpp

Removed: 
    


################################################################################
diff  --git a/clang/include/clang/Basic/AArch64SVEACLETypes.def b/clang/include/clang/Basic/AArch64SVEACLETypes.def
index 56e6179a664e26..55ed9c36f6c5cd 100644
--- a/clang/include/clang/Basic/AArch64SVEACLETypes.def
+++ b/clang/include/clang/Basic/AArch64SVEACLETypes.def
@@ -46,9 +46,8 @@
 //
 //  - ElBits is the size of one element in bits.
 //
-//  - NF enumerates the number of sub-vectors.
-//    TODO: Tuple types are represented as a concatination of "NumEls x ElBits"
-//    vectors.  This will be changed to become a struct containing NF vectors.
+//  - NF enumerates the number of vectors whereby 1 implies a single vector,
+//    with other values implying a struct of NF "NumEls x NumEls" vectors.
 //
 //  - IsSigned is true for vectors of signed integer elements and
 //    for vectors of floating-point elements.

diff  --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 27abeba92999b3..a52e880a764252 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -9990,31 +9990,6 @@ Value *CodeGenFunction::EmitSVEStructLoad(const SVETypeFlags &TypeFlags,
                                           SmallVectorImpl<Value*> &Ops,
                                           unsigned IntID) {
   llvm::ScalableVectorType *VTy = getSVEType(TypeFlags);
-
-  unsigned N;
-  switch (IntID) {
-  case Intrinsic::aarch64_sve_ld2_sret:
-  case Intrinsic::aarch64_sve_ld1_pn_x2:
-  case Intrinsic::aarch64_sve_ldnt1_pn_x2:
-  case Intrinsic::aarch64_sve_ld2q_sret:
-    N = 2;
-    break;
-  case Intrinsic::aarch64_sve_ld3_sret:
-  case Intrinsic::aarch64_sve_ld3q_sret:
-    N = 3;
-    break;
-  case Intrinsic::aarch64_sve_ld4_sret:
-  case Intrinsic::aarch64_sve_ld1_pn_x4:
-  case Intrinsic::aarch64_sve_ldnt1_pn_x4:
-  case Intrinsic::aarch64_sve_ld4q_sret:
-    N = 4;
-    break;
-  default:
-    llvm_unreachable("unknown intrinsic!");
-  }
-  auto RetTy = llvm::VectorType::get(VTy->getElementType(),
-                                     VTy->getElementCount() * N);
-
   Value *Predicate = EmitSVEPredicateCast(Ops[0], VTy);
   Value *BasePtr = Ops[1];
 
@@ -10023,15 +9998,7 @@ Value *CodeGenFunction::EmitSVEStructLoad(const SVETypeFlags &TypeFlags,
     BasePtr = Builder.CreateGEP(VTy, BasePtr, Ops[2]);
 
   Function *F = CGM.getIntrinsic(IntID, {VTy});
-  Value *Call = Builder.CreateCall(F, {Predicate, BasePtr});
-  unsigned MinElts = VTy->getMinNumElements();
-  Value *Ret = llvm::PoisonValue::get(RetTy);
-  for (unsigned I = 0; I < N; I++) {
-    Value *Idx = ConstantInt::get(CGM.Int64Ty, I * MinElts);
-    Value *SRet = Builder.CreateExtractValue(Call, I);
-    Ret = Builder.CreateInsertVector(RetTy, Ret, SRet, Idx);
-  }
-  return Ret;
+  return Builder.CreateCall(F, {Predicate, BasePtr});
 }
 
 Value *CodeGenFunction::EmitSVEStructStore(const SVETypeFlags &TypeFlags,
@@ -10304,6 +10271,19 @@ Value *CodeGenFunction::EmitSVEReinterpret(Value *Val, llvm::Type *Ty) {
   // view (when storing/reloading), whereas the svreinterpret builtin
   // implements bitwise equivalent cast from register point of view.
   // LLVM CodeGen for a bitcast must add an explicit REV for big-endian.
+
+  if (auto *StructTy = dyn_cast<StructType>(Ty)) {
+    Value *Tuple = llvm::PoisonValue::get(Ty);
+
+    for (unsigned I = 0; I < StructTy->getNumElements(); ++I) {
+      Value *In = Builder.CreateExtractValue(Val, I);
+      Value *Out = Builder.CreateBitCast(In, StructTy->getTypeAtIndex(I));
+      Tuple = Builder.CreateInsertValue(Tuple, Out, I);
+    }
+
+    return Tuple;
+  }
+
   return Builder.CreateBitCast(Val, Ty);
 }
 
@@ -10346,44 +10326,26 @@ CodeGenFunction::getSVEOverloadTypes(const SVETypeFlags &TypeFlags,
 }
 
 Value *CodeGenFunction::EmitSVETupleSetOrGet(const SVETypeFlags &TypeFlags,
-                                             llvm::Type *Ty,
                                              ArrayRef<Value *> Ops) {
   assert((TypeFlags.isTupleSet() || TypeFlags.isTupleGet()) &&
          "Expects TypleFlags.isTupleSet() or TypeFlags.isTupleGet()");
-
-  unsigned I = cast<ConstantInt>(Ops[1])->getSExtValue();
-  auto *SingleVecTy = dyn_cast<llvm::ScalableVectorType>(
-      TypeFlags.isTupleSet() ? Ops[2]->getType() : Ty);
-
-  if (!SingleVecTy)
-    return nullptr;
-
-  Value *Idx = ConstantInt::get(CGM.Int64Ty,
-                                I * SingleVecTy->getMinNumElements());
+  unsigned Idx = cast<ConstantInt>(Ops[1])->getZExtValue();
 
   if (TypeFlags.isTupleSet())
-    return Builder.CreateInsertVector(Ty, Ops[0], Ops[2], Idx);
-  return Builder.CreateExtractVector(Ty, Ops[0], Idx);
+    return Builder.CreateInsertValue(Ops[0], Ops[2], Idx);
+  return Builder.CreateExtractValue(Ops[0], Idx);
 }
 
 Value *CodeGenFunction::EmitSVETupleCreate(const SVETypeFlags &TypeFlags,
-                                             llvm::Type *Ty,
-                                             ArrayRef<Value *> Ops) {
+                                           llvm::Type *Ty,
+                                           ArrayRef<Value *> Ops) {
   assert(TypeFlags.isTupleCreate() && "Expects TypleFlag isTupleCreate");
 
-  auto *SrcTy = dyn_cast<llvm::ScalableVectorType>(Ops[0]->getType());
-
-  if (!SrcTy)
-    return nullptr;
+  Value *Tuple = llvm::PoisonValue::get(Ty);
+  for (unsigned Idx = 0; Idx < Ops.size(); Idx++)
+    Tuple = Builder.CreateInsertValue(Tuple, Ops[Idx], Idx);
 
-  unsigned MinElts = SrcTy->getMinNumElements();
-  Value *Call = llvm::PoisonValue::get(Ty);
-  for (unsigned I = 0; I < Ops.size(); I++) {
-    Value *Idx = ConstantInt::get(CGM.Int64Ty, I * MinElts);
-    Call = Builder.CreateInsertVector(Ty, Call, Ops[I], Idx);
-  }
-
-  return Call;
+  return Tuple;
 }
 
 Value *CodeGenFunction::FormSVEBuiltinResult(Value *Call) {
@@ -10453,27 +10415,14 @@ void CodeGenFunction::GetAArch64SVEProcessedOperands(
       continue;
     }
 
-    if (IsTupleGetOrSet || !isa<ScalableVectorType>(Arg->getType())) {
-      Ops.push_back(Arg);
-      continue;
-    }
+    if (isa<StructType>(Arg->getType()) && !IsTupleGetOrSet) {
+      for (unsigned I = 0; I < Arg->getType()->getStructNumElements(); ++I)
+        Ops.push_back(Builder.CreateExtractValue(Arg, I));
 
-    auto *VTy = cast<ScalableVectorType>(Arg->getType());
-    unsigned MinElts = VTy->getMinNumElements();
-    bool IsPred = VTy->getElementType()->isIntegerTy(1);
-    unsigned N = (MinElts * VTy->getScalarSizeInBits()) / (IsPred ? 16 : 128);
-
-    if (N == 1) {
-      Ops.push_back(Arg);
       continue;
     }
 
-    for (unsigned I = 0; I < N; ++I) {
-      Value *Idx = ConstantInt::get(CGM.Int64Ty, (I * MinElts) / N);
-      auto *NewVTy =
-          ScalableVectorType::get(VTy->getElementType(), MinElts / N);
-      Ops.push_back(Builder.CreateExtractVector(NewVTy, Arg, Idx));
-    }
+    Ops.push_back(Arg);
   }
 }
 
@@ -10511,7 +10460,7 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
   else if (TypeFlags.isStructStore())
     return EmitSVEStructStore(TypeFlags, Ops, Builtin->LLVMIntrinsic);
   else if (TypeFlags.isTupleSet() || TypeFlags.isTupleGet())
-    return EmitSVETupleSetOrGet(TypeFlags, Ty, Ops);
+    return EmitSVETupleSetOrGet(TypeFlags, Ops);
   else if (TypeFlags.isTupleCreate())
     return EmitSVETupleCreate(TypeFlags, Ty, Ops);
   else if (TypeFlags.isUndef())

diff  --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
index 9f868c98458996..6802dc7f0c1598 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -4628,7 +4628,6 @@ class CodeGenFunction : public CodeGenTypeCache {
   llvm::ScalableVectorType *getSVEType(const SVETypeFlags &TypeFlags);
   llvm::ScalableVectorType *getSVEPredType(const SVETypeFlags &TypeFlags);
   llvm::Value *EmitSVETupleSetOrGet(const SVETypeFlags &TypeFlags,
-                                    llvm::Type *ReturnType,
                                     ArrayRef<llvm::Value *> Ops);
   llvm::Value *EmitSVETupleCreate(const SVETypeFlags &TypeFlags,
                                   llvm::Type *ReturnType,

diff  --git a/clang/lib/CodeGen/CodeGenTypes.cpp b/clang/lib/CodeGen/CodeGenTypes.cpp
index 5eebd8ad2a0653..d087e714882b33 100644
--- a/clang/lib/CodeGen/CodeGenTypes.cpp
+++ b/clang/lib/CodeGen/CodeGenTypes.cpp
@@ -509,9 +509,20 @@ llvm::Type *CodeGenTypes::ConvertType(QualType T) {
       {
         ASTContext::BuiltinVectorTypeInfo Info =
             Context.getBuiltinVectorTypeInfo(cast<BuiltinType>(Ty));
-        return llvm::ScalableVectorType::get(ConvertType(Info.ElementType),
-                                             Info.EC.getKnownMinValue() *
-                                                 Info.NumVectors);
+        auto VTy =
+            llvm::VectorType::get(ConvertType(Info.ElementType), Info.EC);
+        switch (Info.NumVectors) {
+        default:
+          llvm_unreachable("Expected 1, 2, 3 or 4 vectors!");
+        case 1:
+          return VTy;
+        case 2:
+          return llvm::StructType::get(VTy, VTy);
+        case 3:
+          return llvm::StructType::get(VTy, VTy, VTy);
+        case 4:
+          return llvm::StructType::get(VTy, VTy, VTy, VTy);
+        }
       }
     case BuiltinType::SveCount:
       return llvm::TargetExtType::get(getLLVMContext(), "aarch64.svcount");

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c
index 135bcf743e5e6a..7e7597f82136ca 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add.c
@@ -24,16 +24,12 @@
 // x2
 // CHECK-LABEL: @test_svadd_write_single2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single2_s32j11svint32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single2_s32(uint32_t slice_base, svint32x2_t zn, svint32_t zm) __arm_streaming __arm_inout("za") {
@@ -42,16 +38,12 @@ void test_svadd_write_single2_s32(uint32_t slice_base, svint32x2_t zn, svint32_t
 
 // CHECK-LABEL: @test_svadd_write_single2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single2_u32j12svuint32x2_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32_t zm) __arm_streaming __arm_inout("za") {
@@ -60,16 +52,12 @@ void test_svadd_write_single2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32
 
 // CHECK-LABEL: @test_svadd_write_single2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single2_s64j11svint64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single2_s64(uint32_t slice_base, svint64x2_t zn, svint64_t zm) __arm_streaming __arm_inout("za") {
@@ -78,16 +66,12 @@ void test_svadd_write_single2_s64(uint32_t slice_base, svint64x2_t zn, svint64_t
 
 // CHECK-LABEL: @test_svadd_write_single2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single2_u64j12svuint64x2_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64_t zm) __arm_streaming __arm_inout("za") {
@@ -98,20 +82,12 @@ void test_svadd_write_single2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64
 
 // CHECK-LABEL: @test_svadd_write_single4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single4_s32j11svint32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single4_s32(uint32_t slice_base, svint32x4_t zn, svint32_t zm) __arm_streaming __arm_inout("za") {
@@ -120,20 +96,12 @@ void test_svadd_write_single4_s32(uint32_t slice_base, svint32x4_t zn, svint32_t
 
 // CHECK-LABEL: @test_svadd_write_single4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single4_u32j12svuint32x4_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32_t zm) __arm_streaming __arm_inout("za") {
@@ -142,20 +110,12 @@ void test_svadd_write_single4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32
 
 // CHECK-LABEL: @test_svadd_write_single4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single4_s64j11svint64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single4_s64(uint32_t slice_base, svint64x4_t zn, svint64_t zm) __arm_streaming __arm_inout("za") {
@@ -164,20 +124,12 @@ void test_svadd_write_single4_s64(uint32_t slice_base, svint64x4_t zn, svint64_t
 
 // CHECK-LABEL: @test_svadd_write_single4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_write_single4_u64j12svuint64x4_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_single4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64_t zm) __arm_streaming __arm_inout("za") {
@@ -192,20 +144,12 @@ void test_svadd_write_single4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64
 
 // CHECK-LABEL: @test_svadd_write_multi2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi2_s32j11svint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi2_s32(uint32_t slice_base, svint32x2_t zn, svint32x2_t zm) __arm_streaming __arm_inout("za") {
@@ -214,20 +158,12 @@ void test_svadd_write_multi2_s32(uint32_t slice_base, svint32x2_t zn, svint32x2_
 
 // CHECK-LABEL: @test_svadd_write_multi2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi2_u32j12svuint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32x2_t zm) __arm_streaming __arm_inout("za") {
@@ -236,20 +172,12 @@ void test_svadd_write_multi2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32x
 
 // CHECK-LABEL: @test_svadd_write_multi2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi2_s64j11svint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi2_s64(uint32_t slice_base, svint64x2_t zn, svint64x2_t zm) __arm_streaming __arm_inout("za") {
@@ -258,20 +186,12 @@ void test_svadd_write_multi2_s64(uint32_t slice_base, svint64x2_t zn, svint64x2_
 
 // CHECK-LABEL: @test_svadd_write_multi2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi2_u64j12svuint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64x2_t zm) __arm_streaming __arm_inout("za") {
@@ -282,28 +202,12 @@ void test_svadd_write_multi2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64x
 
 // CHECK-LABEL: @test_svadd_write_multi4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi4_s32j11svint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi4_s32(uint32_t slice_base, svint32x4_t zn, svint32x4_t zm) __arm_streaming __arm_inout("za") {
@@ -312,28 +216,12 @@ void test_svadd_write_multi4_s32(uint32_t slice_base, svint32x4_t zn, svint32x4_
 
 // CHECK-LABEL: @test_svadd_write_multi4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi4_u32j12svuint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32x4_t zm) __arm_streaming __arm_inout("za") {
@@ -342,28 +230,12 @@ void test_svadd_write_multi4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32x
 
 // CHECK-LABEL: @test_svadd_write_multi4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi4_s64j11svint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi4_s64(uint32_t slice_base, svint64x4_t zn, svint64x4_t zm) __arm_streaming __arm_inout("za") {
@@ -372,28 +244,12 @@ void test_svadd_write_multi4_s64(uint32_t slice_base, svint64x4_t zn, svint64x4_
 
 // CHECK-LABEL: @test_svadd_write_multi4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svadd_write_multi4_u64j12svuint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_write_multi4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64x4_t zm) __arm_streaming __arm_inout("za") {
@@ -408,16 +264,12 @@ void test_svadd_write_multi4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64x
 
 // CHECK-LABEL: @test_svadd_za32_vg1x2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za32_vg1x2_f32j13svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za32_vg1x2_f32(uint32_t slice_base, svfloat32x2_t zn) __arm_streaming __arm_inout("za") {
@@ -426,16 +278,12 @@ void test_svadd_za32_vg1x2_f32(uint32_t slice_base, svfloat32x2_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svadd_za32_vg1x2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za32_vg1x2_s32j11svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za32_vg1x2_s32(uint32_t slice_base, svint32x2_t zn) __arm_streaming __arm_inout("za") {
@@ -444,16 +292,12 @@ void test_svadd_za32_vg1x2_s32(uint32_t slice_base, svint32x2_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svadd_za32_vg1x2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za32_vg1x2_u32j12svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za32_vg1x2_u32(uint32_t slice_base, svuint32x2_t zn) __arm_streaming __arm_inout("za") {
@@ -462,16 +306,12 @@ void test_svadd_za32_vg1x2_u32(uint32_t slice_base, svuint32x2_t zn) __arm_strea
 
 // CHECK-LABEL: @test_svadd_za64_vg1x2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za64_vg1x2_f64j13svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za64_vg1x2_f64(uint32_t slice_base, svfloat64x2_t zn) __arm_streaming __arm_inout("za") {
@@ -480,16 +320,12 @@ void test_svadd_za64_vg1x2_f64(uint32_t slice_base, svfloat64x2_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svadd_za64_vg1x2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za64_vg1x2_s64j11svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za64_vg1x2_s64(uint32_t slice_base, svint64x2_t zn) __arm_streaming __arm_inout("za") {
@@ -498,16 +334,12 @@ void test_svadd_za64_vg1x2_s64(uint32_t slice_base, svint64x2_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svadd_za64_vg1x2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za64_vg1x2_u64j12svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za64_vg1x2_u64(uint32_t slice_base, svuint64x2_t zn) __arm_streaming __arm_inout("za") {
@@ -518,20 +350,12 @@ void test_svadd_za64_vg1x2_u64(uint32_t slice_base, svuint64x2_t zn) __arm_strea
 
 // CHECK-LABEL: @test_svadd_za32_vg1x4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za32_vg1x4_f32j13svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za32_vg1x4_f32(uint32_t slice_base, svfloat32x4_t zn) __arm_streaming __arm_inout("za") {
@@ -540,20 +364,12 @@ void test_svadd_za32_vg1x4_f32(uint32_t slice_base, svfloat32x4_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svadd_za32_vg1x4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za32_vg1x4_s32j11svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za32_vg1x4_s32(uint32_t slice_base, svint32x4_t zn) __arm_streaming __arm_inout("za") {
@@ -562,20 +378,12 @@ void test_svadd_za32_vg1x4_s32(uint32_t slice_base, svint32x4_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svadd_za32_vg1x4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za32_vg1x4_u32j12svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za32_vg1x4_u32(uint32_t slice_base, svuint32x4_t zn) __arm_streaming __arm_inout("za") {
@@ -584,20 +392,12 @@ void test_svadd_za32_vg1x4_u32(uint32_t slice_base, svuint32x4_t zn) __arm_strea
 
 // CHECK-LABEL: @test_svadd_za64_vg1x4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za64_vg1x4_f64j13svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za64_vg1x4_f64(uint32_t slice_base, svfloat64x4_t zn) __arm_streaming __arm_inout("za") {
@@ -606,20 +406,12 @@ void test_svadd_za64_vg1x4_f64(uint32_t slice_base, svfloat64x4_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svadd_za64_vg1x4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za64_vg1x4_s64j11svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za64_vg1x4_s64(uint32_t slice_base, svint64x4_t zn) __arm_streaming __arm_inout("za") {
@@ -628,20 +420,12 @@ void test_svadd_za64_vg1x4_s64(uint32_t slice_base, svint64x4_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svadd_za64_vg1x4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svadd_za64_vg1x4_u64j12svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svadd_za64_vg1x4_u64(uint32_t slice_base, svuint64x4_t zn) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
index c88c6978719115..6bd9eab5f1846b 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
@@ -17,19 +17,15 @@
 #endif
 
 // CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_f16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x2_f16j13svfloat16x2_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
@@ -37,23 +33,15 @@ void test_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming
 }
 
 // CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_f16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x4_f16j13svfloat16x4_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
@@ -61,19 +49,15 @@ void test_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming
 }
 
 // CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_f16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x2_f16j13svfloat16x2_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
@@ -81,23 +65,15 @@ void test_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming
 }
 
 // CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_f16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x4_f16j13svfloat16x4_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
@@ -105,19 +81,15 @@ void test_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming
 }
 
 // CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_bf16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x2_bf16j14svbfloat16x2_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svadd_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
@@ -125,23 +97,15 @@ void test_svadd_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streami
 }
 
 // CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_bf16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x4_bf16j14svbfloat16x4_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svadd_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {
@@ -149,19 +113,15 @@ void test_svadd_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streami
 }
 
 // CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_bf16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x2_bf16j14svbfloat16x2_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svsub_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
@@ -169,23 +129,15 @@ void test_svsub_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streami
 }
 
 // CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_bf16(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x4_bf16j14svbfloat16x4_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svsub_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_clamp.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_clamp.c
index 972a6582998838..9c639984305d1b 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_clamp.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_clamp.c
@@ -24,25 +24,27 @@
 
 // CHECK-LABEL: @test_svclamp_single_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svclamp_single_s8_x210svint8x2_tu10__SVInt8_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svclamp_single_s8_x2(svint8x2_t op1, svint8_t op2, svint8_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s8_x2, , )(op1, op2, op3);
@@ -50,25 +52,27 @@ svint8x2_t test_svclamp_single_s8_x2(svint8x2_t op1, svint8_t op2, svint8_t op3)
 
 // CHECK-LABEL: @test_svclamp_single_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_s16_x211svint16x2_tu11__SVInt16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svclamp_single_s16_x2(svint16x2_t op1, svint16_t op2, svint16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s16_x2, , )(op1, op2, op3);
@@ -76,25 +80,27 @@ svint16x2_t test_svclamp_single_s16_x2(svint16x2_t op1, svint16_t op2, svint16_t
 
 // CHECK-LABEL: @test_svclamp_single_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_s32_x211svint32x2_tu11__SVInt32_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svclamp_single_s32_x2(svint32x2_t op1, svint32_t op2, svint32_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s32_x2, , )(op1, op2, op3);
@@ -102,25 +108,27 @@ svint32x2_t test_svclamp_single_s32_x2(svint32x2_t op1, svint32_t op2, svint32_t
 
 // CHECK-LABEL: @test_svclamp_single_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_s64_x211svint64x2_tu11__SVInt64_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svclamp_single_s64_x2(svint64x2_t op1, svint64_t op2, svint64_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s64_x2, , )(op1, op2, op3);
@@ -131,37 +139,35 @@ svint64x2_t test_svclamp_single_s64_x2(svint64x2_t op1, svint64_t op2, svint64_t
 
 // CHECK-LABEL: @test_svclamp_single_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP1_COERCE2:%.*]], <vscale x 16 x i8> [[OP1_COERCE3:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svclamp_single_s8_x410svint8x4_tu10__SVInt8_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP1_COERCE2:%.*]], <vscale x 16 x i8> [[OP1_COERCE3:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svclamp_single_s8_x4(svint8x4_t op1, svint8_t op2, svint8_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s8_x4, , )(op1, op2, op3);
@@ -169,37 +175,35 @@ svint8x4_t test_svclamp_single_s8_x4(svint8x4_t op1, svint8_t op2, svint8_t op3)
 
 // CHECK-LABEL: @test_svclamp_single_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP1_COERCE2:%.*]], <vscale x 8 x i16> [[OP1_COERCE3:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_s16_x411svint16x4_tu11__SVInt16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP1_COERCE2:%.*]], <vscale x 8 x i16> [[OP1_COERCE3:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svclamp_single_s16_x4(svint16x4_t op1, svint16_t op2, svint16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s16_x4, , )(op1, op2, op3);
@@ -207,37 +211,35 @@ svint16x4_t test_svclamp_single_s16_x4(svint16x4_t op1, svint16_t op2, svint16_t
 
 // CHECK-LABEL: @test_svclamp_single_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP1_COERCE2:%.*]], <vscale x 4 x i32> [[OP1_COERCE3:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_s32_x411svint32x4_tu11__SVInt32_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP1_COERCE2:%.*]], <vscale x 4 x i32> [[OP1_COERCE3:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svclamp_single_s32_x4(svint32x4_t op1, svint32_t op2, svint32_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s32_x4, , )(op1, op2, op3);
@@ -245,37 +247,35 @@ svint32x4_t test_svclamp_single_s32_x4(svint32x4_t op1, svint32_t op2, svint32_t
 
 // CHECK-LABEL: @test_svclamp_single_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP1_COERCE2:%.*]], <vscale x 2 x i64> [[OP1_COERCE3:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_s64_x411svint64x4_tu11__SVInt64_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP1_COERCE2:%.*]], <vscale x 2 x i64> [[OP1_COERCE3:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svclamp_single_s64_x4(svint64x4_t op1, svint64_t op2, svint64_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_s64_x4, , )(op1, op2, op3);
@@ -288,25 +288,27 @@ svint64x4_t test_svclamp_single_s64_x4(svint64x4_t op1, svint64_t op2, svint64_t
 
 // CHECK-LABEL: @test_svclamp_single_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svclamp_single_u8_x211svuint8x2_tu11__SVUint8_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x2.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svclamp_single_u8_x2(svuint8x2_t op1, svuint8_t op2, svuint8_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u8_x2, , )(op1, op2, op3);
@@ -314,25 +316,27 @@ svuint8x2_t test_svclamp_single_u8_x2(svuint8x2_t op1, svuint8_t op2, svuint8_t
 
 // CHECK-LABEL: @test_svclamp_single_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_u16_x212svuint16x2_tu12__SVUint16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x2.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svclamp_single_u16_x2(svuint16x2_t op1, svuint16_t op2, svuint16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u16_x2, , )(op1, op2, op3);
@@ -340,25 +344,27 @@ svuint16x2_t test_svclamp_single_u16_x2(svuint16x2_t op1, svuint16_t op2, svuint
 
 // CHECK-LABEL: @test_svclamp_single_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_u32_x212svuint32x2_tu12__SVUint32_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x2.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svclamp_single_u32_x2(svuint32x2_t op1, svuint32_t op2, svuint32_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u32_x2, , )(op1, op2, op3);
@@ -366,25 +372,27 @@ svuint32x2_t test_svclamp_single_u32_x2(svuint32x2_t op1, svuint32_t op2, svuint
 
 // CHECK-LABEL: @test_svclamp_single_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_u64_x212svuint64x2_tu12__SVUint64_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x2.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svclamp_single_u64_x2(svuint64x2_t op1, svuint64_t op2, svuint64_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u64_x2, , )(op1, op2, op3);
@@ -395,37 +403,35 @@ svuint64x2_t test_svclamp_single_u64_x2(svuint64x2_t op1, svuint64_t op2, svuint
 
 // CHECK-LABEL: @test_svclamp_single_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP1_COERCE2:%.*]], <vscale x 16 x i8> [[OP1_COERCE3:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svclamp_single_u8_x411svuint8x4_tu11__SVUint8_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[OP1]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uclamp.single.x4.nxv16i8(<vscale x 16 x i8> [[OP1_COERCE0:%.*]], <vscale x 16 x i8> [[OP1_COERCE1:%.*]], <vscale x 16 x i8> [[OP1_COERCE2:%.*]], <vscale x 16 x i8> [[OP1_COERCE3:%.*]], <vscale x 16 x i8> [[OP2:%.*]], <vscale x 16 x i8> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svclamp_single_u8_x4(svuint8x4_t op1, svuint8_t op2, svuint8_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u8_x4, , )(op1, op2, op3);
@@ -433,37 +439,35 @@ svuint8x4_t test_svclamp_single_u8_x4(svuint8x4_t op1, svuint8_t op2, svuint8_t
 
 // CHECK-LABEL: @test_svclamp_single_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP1_COERCE2:%.*]], <vscale x 8 x i16> [[OP1_COERCE3:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_u16_x412svuint16x4_tu12__SVUint16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uclamp.single.x4.nxv8i16(<vscale x 8 x i16> [[OP1_COERCE0:%.*]], <vscale x 8 x i16> [[OP1_COERCE1:%.*]], <vscale x 8 x i16> [[OP1_COERCE2:%.*]], <vscale x 8 x i16> [[OP1_COERCE3:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svclamp_single_u16_x4(svuint16x4_t op1, svuint16_t op2, svuint16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u16_x4, , )(op1, op2, op3);
@@ -471,37 +475,35 @@ svuint16x4_t test_svclamp_single_u16_x4(svuint16x4_t op1, svuint16_t op2, svuint
 
 // CHECK-LABEL: @test_svclamp_single_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP1_COERCE2:%.*]], <vscale x 4 x i32> [[OP1_COERCE3:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_u32_x412svuint32x4_tu12__SVUint32_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uclamp.single.x4.nxv4i32(<vscale x 4 x i32> [[OP1_COERCE0:%.*]], <vscale x 4 x i32> [[OP1_COERCE1:%.*]], <vscale x 4 x i32> [[OP1_COERCE2:%.*]], <vscale x 4 x i32> [[OP1_COERCE3:%.*]], <vscale x 4 x i32> [[OP2:%.*]], <vscale x 4 x i32> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svclamp_single_u32_x4(svuint32x4_t op1, svuint32_t op2, svuint32_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u32_x4, , )(op1, op2, op3);
@@ -509,37 +511,35 @@ svuint32x4_t test_svclamp_single_u32_x4(svuint32x4_t op1, svuint32_t op2, svuint
 
 // CHECK-LABEL: @test_svclamp_single_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP1_COERCE2:%.*]], <vscale x 2 x i64> [[OP1_COERCE3:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_u64_x412svuint64x4_tu12__SVUint64_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uclamp.single.x4.nxv2i64(<vscale x 2 x i64> [[OP1_COERCE0:%.*]], <vscale x 2 x i64> [[OP1_COERCE1:%.*]], <vscale x 2 x i64> [[OP1_COERCE2:%.*]], <vscale x 2 x i64> [[OP1_COERCE3:%.*]], <vscale x 2 x i64> [[OP2:%.*]], <vscale x 2 x i64> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svclamp_single_u64_x4(svuint64x4_t op1, svuint64_t op2, svuint64_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_u64_x4, , )(op1, op2, op3);
@@ -552,25 +552,27 @@ svuint64x4_t test_svclamp_single_u64_x4(svuint64x4_t op1, svuint64_t op2, svuint
 
 // CHECK-LABEL: @test_svclamp_single_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x2.nxv8f16(<vscale x 8 x half> [[OP1_COERCE0:%.*]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_f16_x213svfloat16x2_tu13__SVFloat16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x2.nxv8f16(<vscale x 8 x half> [[OP1_COERCE0:%.*]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svclamp_single_f16_x2(svfloat16x2_t op1, svfloat16_t op2, svfloat16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_f16_x2, , )(op1, op2, op3);
@@ -578,25 +580,27 @@ svfloat16x2_t test_svclamp_single_f16_x2(svfloat16x2_t op1, svfloat16_t op2, svf
 
 // CHECK-LABEL: @test_svclamp_single_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x2.nxv4f32(<vscale x 4 x float> [[OP1_COERCE0:%.*]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_f32_x213svfloat32x2_tu13__SVFloat32_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x2.nxv4f32(<vscale x 4 x float> [[OP1_COERCE0:%.*]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svclamp_single_f32_x2(svfloat32x2_t op1, svfloat32_t op2, svfloat32_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_f32_x2, , )(op1, op2, op3);
@@ -605,25 +609,27 @@ svfloat32x2_t test_svclamp_single_f32_x2(svfloat32x2_t op1, svfloat32_t op2, svf
 
 // CHECK-LABEL: @test_svclamp_single_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x2.nxv2f64(<vscale x 2 x double> [[OP1_COERCE0:%.*]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_f64_x213svfloat64x2_tu13__SVFloat64_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x2.nxv2f64(<vscale x 2 x double> [[OP1_COERCE0:%.*]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svclamp_single_f64_x2(svfloat64x2_t op1, svfloat64_t op2, svfloat64_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_f64_x2, , )(op1, op2, op3);
@@ -634,37 +640,35 @@ svfloat64x2_t test_svclamp_single_f64_x2(svfloat64x2_t op1, svfloat64_t op2, svf
 
 // CHECK-LABEL: @test_svclamp_single_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x4.nxv8f16(<vscale x 8 x half> [[OP1_COERCE0:%.*]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], <vscale x 8 x half> [[OP1_COERCE2:%.*]], <vscale x 8 x half> [[OP1_COERCE3:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_f16_x413svfloat16x4_tu13__SVFloat16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fclamp.single.x4.nxv8f16(<vscale x 8 x half> [[OP1_COERCE0:%.*]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], <vscale x 8 x half> [[OP1_COERCE2:%.*]], <vscale x 8 x half> [[OP1_COERCE3:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svclamp_single_f16_x4(svfloat16x4_t op1, svfloat16_t op2, svfloat16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_f16_x4, , )(op1, op2, op3);
@@ -672,37 +676,35 @@ svfloat16x4_t test_svclamp_single_f16_x4(svfloat16x4_t op1, svfloat16_t op2, svf
 
 // CHECK-LABEL: @test_svclamp_single_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x4.nxv4f32(<vscale x 4 x float> [[OP1_COERCE0:%.*]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], <vscale x 4 x float> [[OP1_COERCE2:%.*]], <vscale x 4 x float> [[OP1_COERCE3:%.*]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_f32_x413svfloat32x4_tu13__SVFloat32_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fclamp.single.x4.nxv4f32(<vscale x 4 x float> [[OP1_COERCE0:%.*]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], <vscale x 4 x float> [[OP1_COERCE2:%.*]], <vscale x 4 x float> [[OP1_COERCE3:%.*]], <vscale x 4 x float> [[OP2:%.*]], <vscale x 4 x float> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svclamp_single_f32_x4(svfloat32x4_t op1, svfloat32_t op2, svfloat32_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_f32_x4, , )(op1, op2, op3);
@@ -710,37 +712,35 @@ svfloat32x4_t test_svclamp_single_f32_x4(svfloat32x4_t op1, svfloat32_t op2, svf
 
 // CHECK-LABEL: @test_svclamp_single_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x4.nxv2f64(<vscale x 2 x double> [[OP1_COERCE0:%.*]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], <vscale x 2 x double> [[OP1_COERCE2:%.*]], <vscale x 2 x double> [[OP1_COERCE3:%.*]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svclamp_single_f64_x413svfloat64x4_tu13__SVFloat64_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fclamp.single.x4.nxv2f64(<vscale x 2 x double> [[OP1_COERCE0:%.*]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], <vscale x 2 x double> [[OP1_COERCE2:%.*]], <vscale x 2 x double> [[OP1_COERCE3:%.*]], <vscale x 2 x double> [[OP2:%.*]], <vscale x 2 x double> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svclamp_single_f64_x4(svfloat64x4_t op1, svfloat64_t op2, svfloat64_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_f64_x4, , )(op1, op2, op3);
@@ -748,25 +748,27 @@ svfloat64x4_t test_svclamp_single_f64_x4(svfloat64x4_t op1, svfloat64_t op2, svf
 
 // CHECK-LABEL: @test_svclamp_single_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[OP1_COERCE0:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE1:%.*]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svclamp_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[OP1_COERCE0:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE1:%.*]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svclamp_single_bf16_x2(svbfloat16x2_t op1, svbfloat16_t op2, svbfloat16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_bf16_x2, , )(op1, op2, op3);
@@ -774,37 +776,35 @@ svbfloat16x2_t test_svclamp_single_bf16_x2(svbfloat16x2_t op1, svbfloat16_t op2,
 
 // CHECK-LABEL: @test_svclamp_single_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[OP1_COERCE0:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE1:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE2:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE3:%.*]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svclamp_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[OP1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfclamp.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[OP1_COERCE0:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE1:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE2:%.*]], <vscale x 8 x bfloat> [[OP1_COERCE3:%.*]], <vscale x 8 x bfloat> [[OP2:%.*]], <vscale x 8 x bfloat> [[OP3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svclamp_single_bf16_x4(svbfloat16x4_t op1, svbfloat16_t op2, svbfloat16_t op3) __arm_streaming {
   return SVE_ACLE_FUNC(svclamp, _single_bf16_x4, , )(op1, op2, op3);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c
index 686e4f4f99afe9..2d61670fd60493 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c
@@ -19,17 +19,13 @@
 
 // CHECK-LABEL: @test_cvt_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvt.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvt.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z15test_cvt_f16_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvt.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvt.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
 //
 svfloat16_t test_cvt_f16_x2(svfloat32x2_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_f16,_f32_x2,,)(zn);
@@ -37,17 +33,13 @@ svfloat16_t test_cvt_f16_x2(svfloat32x2_t zn)  __arm_streaming {
 
 // CHECK-LABEL: @test_cvt_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvt.x2(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvt.x2(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z16test_cvt_bf16_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvt.x2(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvt.x2(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
 //
 svbfloat16_t test_cvt_bf16_x2(svfloat32x2_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_bf16,_f32_x2,,)(zn);
@@ -57,25 +49,27 @@ svbfloat16_t test_cvt_bf16_x2(svfloat32x2_t zn)  __arm_streaming {
 // x2
 // CHECK-LABEL: @test_svcvt_f32_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_f32_u32_x212svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svcvt_f32_u32_x2(svuint32x2_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_f32,_u32_x2,,)(zn);
@@ -83,25 +77,27 @@ svfloat32x2_t test_svcvt_f32_u32_x2(svuint32x2_t zn)  __arm_streaming {
 
 // CHECK-LABEL: @test_svcvt_f32_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_f32_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x2.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svcvt_f32_s32_x2(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_f32,_s32_x2,,)(zn);
@@ -109,25 +105,27 @@ svfloat32x2_t test_svcvt_f32_s32_x2(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svcvt_u32_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_u32_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svcvt_u32_f32_x2(svfloat32x2_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_u32,_f32_x2,,)(zn);
@@ -135,25 +133,27 @@ svuint32x2_t test_svcvt_u32_f32_x2(svfloat32x2_t zn)  __arm_streaming {
 
 // CHECK-LABEL: @test_svcvt_s32_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_s32_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x2.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svcvt_s32_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_s32,_f32_x2,,)(zn);
@@ -162,37 +162,35 @@ svint32x2_t test_svcvt_s32_f32_x2(svfloat32x2_t zn) __arm_streaming {
 // x4
 // CHECK-LABEL: @test_svcvt_f32_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_f32_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ucvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svcvt_f32_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_f32,_u32_x4,,)(zn);
@@ -200,37 +198,35 @@ svfloat32x4_t test_svcvt_f32_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svcvt_f32_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_f32_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.scvtf.x4.nxv4f32.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svcvt_f32_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_f32,_s32_x4,,)(zn);
@@ -238,37 +234,35 @@ svfloat32x4_t test_svcvt_f32_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svcvt_u32_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_u32_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzu.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svcvt_u32_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_u32,_f32_x4,,)(zn);
@@ -276,37 +270,35 @@ svuint32x4_t test_svcvt_u32_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svcvt_s32_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svcvt_s32_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.fcvtzs.x4.nxv4i32.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svcvt_s32_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_s32,_f32_x4,,)(zn);
@@ -314,17 +306,13 @@ svint32x4_t test_svcvt_s32_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_s16_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvt_s16_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svint16_t test_qcvt_s16_s32_x2(svint32x2_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_s16,_s32_x2,,)(zn);
@@ -332,17 +320,13 @@ svint16_t test_qcvt_s16_s32_x2(svint32x2_t zn)  __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_u16_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvt_u16_u32_x212svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_qcvt_u16_u32_x2(svuint32x2_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_u16,_u32_x2,,)(zn);
@@ -350,17 +334,13 @@ svuint16_t test_qcvt_u16_u32_x2(svuint32x2_t zn)  __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_u16_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvt_u16_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_qcvt_u16_s32_x2(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_u16,_s32_x2,,)(zn);
@@ -368,21 +348,13 @@ svuint16_t test_qcvt_u16_s32_x2(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_u8_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvt.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvt.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z19test_qcvt_u8_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvt.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvt.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_qcvt_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_u8,_u32_x4,,)(zn);
@@ -390,21 +362,13 @@ svuint8_t test_qcvt_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_u16_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvt_u16_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvt.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_qcvt_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_u16,_u64_x4,,)(zn);
@@ -412,21 +376,13 @@ svuint16_t test_qcvt_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_s8_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvt.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvt.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z19test_qcvt_s8_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvt.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvt.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svint8_t test_qcvt_s8_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_s8,_s32_x4,,)(zn);
@@ -434,21 +390,13 @@ svint8_t test_qcvt_s8_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_s16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvt_s16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvt.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svint16_t test_qcvt_s16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_s16,_s64_x4,,)(zn);
@@ -456,21 +404,13 @@ svint16_t test_qcvt_s16_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_u8_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtu.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtu.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z19test_qcvt_u8_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtu.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtu.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_qcvt_u8_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_u8,_s32_x4,,)(zn);
@@ -478,21 +418,13 @@ svuint8_t test_qcvt_u8_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvt_u16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvt_u16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtu.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_qcvt_u16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvt_u16,_s64_x4,,)(zn);
@@ -500,21 +432,27 @@ svuint16_t test_qcvt_u16_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_cvt_f32_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvt.widen.x2.nxv4f32(<vscale x 8 x half> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z15test_cvt_f32_x2u13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvt.widen.x2.nxv4f32(<vscale x 8 x half> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 __attribute__((target("sme-f16f16"))) svfloat32x2_t test_cvt_f32_x2(svfloat16_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_f32,_f16_x2,,)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtl.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtl.c
index 453dd3db6adf09..fc5c0376e925e0 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtl.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtl.c
@@ -19,21 +19,27 @@
 
 // CHECK-LABEL: @test_cvtl_f32_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvtl.widen.x2.nxv4f32(<vscale x 8 x half> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_cvtl_f32_x2u13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvtl.widen.x2.nxv4f32(<vscale x 8 x half> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_cvtl_f32_x2(svfloat16_t zn)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvtl_f32,_f16_x2,,)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtn.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtn.c
index 78aeb67babf4c9..b38bf6fd350846 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtn.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvtn.c
@@ -19,17 +19,13 @@
 
 // CHECK-LABEL: @test_cvtn_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvtn.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvtn.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z16test_cvtn_f16_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvtn.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvtn.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
 //
 svfloat16_t test_cvtn_f16_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvtn_f16,_f32_x2,,)(zn);
@@ -37,17 +33,13 @@ svfloat16_t test_cvtn_f16_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_cvtn_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvtn.x2(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvtn.x2(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z17test_cvtn_bf16_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvtn.x2(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvtn.x2(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
 //
 svbfloat16_t test_cvtn_bf16_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svcvtn_bf16,_f32_x2,,)(zn);
@@ -55,21 +47,13 @@ svbfloat16_t test_cvtn_bf16_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvtn_u8_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvtn_u8_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_qcvtn_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvtn_u8,_u32_x4,,)(zn);
@@ -77,21 +61,13 @@ svuint8_t test_qcvtn_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvtn_u16_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z21test_qcvtn_u16_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_qcvtn_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvtn_u16,_u64_x4,,)(zn);
@@ -99,21 +75,13 @@ svuint16_t test_qcvtn_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvtn_s8_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z20test_qcvtn_s8_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svint8_t test_qcvtn_s8_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvtn_s8,_s32_x4,,)(zn);
@@ -121,21 +89,13 @@ svint8_t test_qcvtn_s8_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvtn_s16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z21test_qcvtn_s16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svint16_t test_qcvtn_s16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvtn_s16,_s64_x4,,)(zn);
@@ -143,21 +103,13 @@ svint16_t test_qcvtn_s16_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvtn_u8_32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtun.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtun.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z19test_qcvtn_u8_32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtun.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqcvtun.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_qcvtn_u8_32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvtn_u8,_s32_x4,,)(zn);
@@ -165,21 +117,13 @@ svuint8_t test_qcvtn_u8_32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_qcvtn_u16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtun.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtun.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z21test_qcvtn_u16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtun.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqcvtun.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_qcvtn_u16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqcvtn_u16,_s64_x4,,)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c
index 5d026f8cde5e05..a1540bba2a8a96 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c
@@ -19,29 +19,27 @@
 
 // CHECK-LABEL: @test_svamax_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamax_f16_x213svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svamax_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamax,_f16_x2)(zdn, zm);
@@ -49,29 +47,27 @@ svfloat16x2_t test_svamax_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamax_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamax_f32_x213svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svamax_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamax,_f32_x2)(zdn, zm);
@@ -79,29 +75,27 @@ svfloat32x2_t test_svamax_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamax_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamax_f64_x213svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svamax_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamax,_f64_x2)(zdn, zm);
@@ -109,29 +103,27 @@ svfloat64x2_t test_svamax_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamin_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamin_f16_x213svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svamin_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamin,_f16_x2)(zdn, zm);
@@ -139,29 +131,27 @@ svfloat16x2_t test_svamin_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamin_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamin_f32_x213svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svamin_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamin,_f32_x2)(zdn, zm);
@@ -169,29 +159,27 @@ svfloat32x2_t test_svamin_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamin_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamin_f64_x213svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svamin_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamin,_f64_x2)(zdn, zm);
@@ -201,45 +189,35 @@ svfloat64x2_t test_svamin_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamax_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamax_f16_x413svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svamax_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamax,_f16_x4)(zdn, zm);
@@ -247,45 +225,35 @@ svfloat16x4_t test_svamax_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamax_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamax_f32_x413svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svamax_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamax,_f32_x4)(zdn, zm);
@@ -293,45 +261,35 @@ svfloat32x4_t test_svamax_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamax_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamax_f64_x413svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svamax_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamax,_f64_x4)(zdn, zm);
@@ -339,45 +297,35 @@ svfloat64x4_t test_svamax_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamin_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamin_f16_x413svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svamin_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamin,_f16_x4)(zdn, zm);
@@ -385,45 +333,35 @@ svfloat16x4_t test_svamin_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamin_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamin_f32_x413svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svamin_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamin,_f32_x4)(zdn, zm);
@@ -431,45 +369,35 @@ svfloat32x4_t test_svamin_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svamin_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svamin_f64_x413svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svamin_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svamin,_f64_x4)(zdn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fmlas16.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fmlas16.c
index 909bf657dec5c7..4338ea030b504f 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fmlas16.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fmlas16.c
@@ -16,19 +16,15 @@
 #endif
 
 // CHECK-LABEL: define dso_local void @test_svmla_single_za16_f16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z32test_svmla_single_za16_f16_vg1x2j13svfloat16x2_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_single_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -36,23 +32,15 @@ void test_svmla_single_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat1
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_single_za16_f16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z32test_svmla_single_za16_f16_vg1x4j13svfloat16x4_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_single_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -60,19 +48,15 @@ void test_svmla_single_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat1
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_single_za16_f16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z32test_svmls_single_za16_f16_vg1x2j13svfloat16x2_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_single_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -80,23 +64,15 @@ void test_svmls_single_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat1
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_single_za16_f16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z32test_svmls_single_za16_f16_vg1x4j13svfloat16x4_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_single_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -104,23 +80,15 @@ void test_svmls_single_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat1
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_za16_f16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 16 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svmla_za16_f16_vg1x2j13svfloat16x2_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 16 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -128,31 +96,15 @@ void test_svmla_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16x2_t z
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_za16_f16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 32 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]], <vscale x 8 x half> [[ZM_COERCE2]], <vscale x 8 x half> [[ZM_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svmla_za16_f16_vg1x4j13svfloat16x4_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 32 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]], <vscale x 8 x half> [[ZM_COERCE2]], <vscale x 8 x half> [[ZM_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -160,23 +112,15 @@ void test_svmla_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16x4_t z
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_za16_f16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 16 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svmls_za16_f16_vg1x2j13svfloat16x2_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 16 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -184,31 +128,15 @@ void test_svmls_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16x2_t z
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_za16_f16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 32 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]], <vscale x 8 x half> [[ZM_COERCE2]], <vscale x 8 x half> [[ZM_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z25test_svmls_za16_f16_vg1x4j13svfloat16x4_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 32 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM_COERCE0]], <vscale x 8 x half> [[ZM_COERCE1]], <vscale x 8 x half> [[ZM_COERCE2]], <vscale x 8 x half> [[ZM_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -216,19 +144,15 @@ void test_svmls_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16x4_t z
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_lane_za16_f16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z30test_svmla_lane_za16_f16_vg1x2j13svfloat16x2_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_lane_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -236,23 +160,15 @@ void test_svmla_lane_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_lane_za16_f16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z30test_svmla_lane_za16_f16_vg1x4j13svfloat16x4_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_lane_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -260,19 +176,15 @@ void test_svmla_lane_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_lane_za16_f16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z30test_svmls_lane_za16_f16_vg1x2j13svfloat16x2_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_lane_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -280,23 +192,15 @@ void test_svmls_lane_za16_f16_vg1x2(uint32_t slice, svfloat16x2_t zn, svfloat16_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_lane_za16_f16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z30test_svmls_lane_za16_f16_vg1x4j13svfloat16x4_tu13__SVFloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[ZN_COERCE0]], <vscale x 8 x half> [[ZN_COERCE1]], <vscale x 8 x half> [[ZN_COERCE2]], <vscale x 8 x half> [[ZN_COERCE3]], <vscale x 8 x half> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_lane_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -304,19 +208,15 @@ void test_svmls_lane_za16_f16_vg1x4(uint32_t slice, svfloat16x4_t zn, svfloat16_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_single_za16_bf16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z33test_svmla_single_za16_bf16_vg1x2j14svbfloat16x2_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_single_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -324,23 +224,15 @@ void test_svmla_single_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbflo
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_single_za16_bf16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z33test_svmla_single_za16_bf16_vg1x4j14svbfloat16x4_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_single_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -348,19 +240,15 @@ void test_svmla_single_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbflo
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_single_za16_bf16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z33test_svmls_single_za16_bf16_vg1x2j14svbfloat16x2_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_single_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -368,23 +256,15 @@ void test_svmls_single_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbflo
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_single_za16_bf16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z33test_svmls_single_za16_bf16_vg1x4j14svbfloat16x4_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_single_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -392,23 +272,15 @@ void test_svmls_single_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbflo
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_za16_bf16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 16 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svmla_za16_bf16_vg1x2j14svbfloat16x2_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 16 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -416,31 +288,15 @@ void test_svmla_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16x2_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_za16_bf16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 32 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE2]], <vscale x 8 x bfloat> [[ZM_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svmla_za16_bf16_vg1x4j14svbfloat16x4_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 32 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE2]], <vscale x 8 x bfloat> [[ZM_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -448,23 +304,15 @@ void test_svmla_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16x4_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_za16_bf16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 16 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svmls_za16_bf16_vg1x2j14svbfloat16x2_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 16 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -472,31 +320,15 @@ void test_svmls_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16x2_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_za16_bf16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 32 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE2]], <vscale x 8 x bfloat> [[ZM_COERCE3]])
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z26test_svmls_za16_bf16_vg1x4j14svbfloat16x4_tS_(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 32 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM_COERCE0]], <vscale x 8 x bfloat> [[ZM_COERCE1]], <vscale x 8 x bfloat> [[ZM_COERCE2]], <vscale x 8 x bfloat> [[ZM_COERCE3]])
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -504,19 +336,15 @@ void test_svmls_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16x4_
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_lane_za16_bf16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z31test_svmla_lane_za16_bf16_vg1x2j14svbfloat16x2_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_lane_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -524,23 +352,15 @@ void test_svmla_lane_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat
 }
 
 // CHECK-LABEL: define dso_local void @test_svmla_lane_za16_bf16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z31test_svmla_lane_za16_bf16_vg1x4j14svbfloat16x4_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmla_lane_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -548,19 +368,15 @@ void test_svmla_lane_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_lane_za16_bf16_vg1x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z31test_svmls_lane_za16_bf16_vg1x2j14svbfloat16x2_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_lane_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -568,23 +384,15 @@ void test_svmls_lane_za16_bf16_vg1x2(uint32_t slice, svbfloat16x2_t zn, svbfloat
 }
 
 // CHECK-LABEL: define dso_local void @test_svmls_lane_za16_bf16_vg1x4(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CHECK-CXX-LABEL: define dso_local void @_Z31test_svmls_lane_za16_bf16_vg1x4j14svbfloat16x4_tu14__SVBfloat16_t(
-// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]]) local_unnamed_addr #[[ATTR0]] {
 // CHECK-CXX-NEXT:  entry:
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
-// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[ZN_COERCE0]], <vscale x 8 x bfloat> [[ZN_COERCE1]], <vscale x 8 x bfloat> [[ZN_COERCE2]], <vscale x 8 x bfloat> [[ZN_COERCE3]], <vscale x 8 x bfloat> [[ZM]], i32 7)
 // CHECK-CXX-NEXT:    ret void
 //
 void test_svmls_lane_za16_bf16_vg1x4(uint32_t slice, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fp_dots.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fp_dots.c
index 3392ba9235bfda..00cbfdbe7ca341 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fp_dots.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_fp_dots.c
@@ -19,20 +19,12 @@
 // Multi, multi (half)
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za32_vg1x2_f16j13svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -41,28 +33,12 @@ void test_svdot_multi_za32_vg1x2_f16(uint32_t slice_base, svfloat16x2_t zn, svfl
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za32_vg1x4_f16j13svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -74,16 +50,12 @@ void test_svdot_multi_za32_vg1x4_f16(uint32_t slice_base, svfloat16x4_t zn, svfl
 // Multi, single (half)
 // CHECK-LABEL: @test_svdot_single_za32_vg1x2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za32_vg1x2_f16j13svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -92,20 +64,12 @@ void test_svdot_single_za32_vg1x2_f16(uint32_t slice_base, svfloat16x2_t zn, svf
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za32_vg1x4_f16j13svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -117,16 +81,12 @@ void test_svdot_single_za32_vg1x4_f16(uint32_t slice_base, svfloat16x4_t zn, svf
 // Multi, indexed (half)
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za32_vg1x2_f16j13svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -135,20 +95,12 @@ void test_svdot_lane_za32_vg1x2_f16(uint32_t slice_base, svfloat16x2_t zn, svflo
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za32_vg1x4_f16j13svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -160,20 +112,12 @@ void test_svdot_lane_za32_vg1x4_f16(uint32_t slice_base, svfloat16x4_t zn, svflo
 // Multi, multi (bfloat)
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_multi_za32_vg1x2_bf16j14svbfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -182,28 +126,12 @@ void test_svdot_multi_za32_vg1x2_bf16(uint32_t slice_base, svbfloat16x2_t zn, sv
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_multi_za32_vg1x4_bf16j14svbfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -215,16 +143,12 @@ void test_svdot_multi_za32_vg1x4_bf16(uint32_t slice_base, svbfloat16x4_t zn, sv
 // Multi, single (bfloat)
 // CHECK-LABEL: @test_svdot_single_za32_vg1x2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z33test_svdot_single_za32_vg1x2_bf16j14svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -233,20 +157,12 @@ void test_svdot_single_za32_vg1x2_bf16(uint32_t slice_base, svbfloat16x2_t zn, s
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z33test_svdot_single_za32_vg1x4_bf16j14svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.single.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -258,16 +174,12 @@ void test_svdot_single_za32_vg1x4_bf16(uint32_t slice_base, svbfloat16x4_t zn, s
 // Multi, indexed (bfloat)
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_lane_za32_vg1x2_bf16j14svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -276,20 +188,12 @@ void test_svdot_lane_za32_vg1x2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svb
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_lane_za32_vg1x4_bf16j14svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fdot.lane.za32.vg1x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_frint.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_frint.c
index 216e44339d8ce6..abdb5a46d54532 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_frint.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_frint.c
@@ -21,25 +21,27 @@
 
 // CHECK-LABEL: @test_svfrinta_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svfrinta_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svfrinta_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrinta,_f32_x2)(zn);
@@ -47,37 +49,35 @@ svfloat32x2_t test_svfrinta_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svfrinta_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svfrinta_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frinta.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svfrinta_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrinta,_f32_x4)(zn);
@@ -87,25 +87,27 @@ svfloat32x4_t test_svfrinta_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svfrintam_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z21test_svfrintam_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svfrintam_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrintm,_f32_x2)(zn);
@@ -113,37 +115,35 @@ svfloat32x2_t test_svfrintam_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svfrintm_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svfrintm_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintm.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svfrintm_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrintm,_f32_x4)(zn);
@@ -153,25 +153,27 @@ svfloat32x4_t test_svfrintm_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svfrintn_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svfrintn_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svfrintn_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrintn,_f32_x2)(zn);
@@ -179,37 +181,35 @@ svfloat32x2_t test_svfrintn_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svfrintn_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svfrintn_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintn.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svfrintn_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrintn,_f32_x4)(zn);
@@ -219,25 +219,27 @@ svfloat32x4_t test_svfrintn_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svfrintp_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svfrintp_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svfrintp_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrintp,_f32_x2)(zn);
@@ -245,37 +247,35 @@ svfloat32x2_t test_svfrintp_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svfrintp_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svfrintp_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.frintp.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svfrintp_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svrintp,_f32_x4)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_int_dots.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_int_dots.c
index ab8b4fc4108c34..961133b89eb29c 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_int_dots.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_int_dots.c
@@ -19,20 +19,12 @@
 // Multi, multi (unsigned)
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za32_vg1x2_u16j12svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -41,28 +33,12 @@ void test_svdot_multi_za32_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuin
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za32_vg1x4_u16j12svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -71,20 +47,12 @@ void test_svdot_multi_za32_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuin
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_multi_za32_vg1x2_u8j11svuint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8x2_t zm) __arm_streaming __arm_inout("za") {
@@ -93,28 +61,12 @@ void test_svdot_multi_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_multi_za32_vg1x4_u8j11svuint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8x4_t zm) __arm_streaming __arm_inout("za") {
@@ -123,20 +75,12 @@ void test_svdot_multi_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8
 
 // CHECK-LABEL: @test_svdot_multi_za64_vg1x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za64_vg1x2_u16j12svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za64_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -145,28 +89,12 @@ void test_svdot_multi_za64_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuin
 
 // CHECK-LABEL: @test_svdot_multi_za64_vg1x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za64_vg1x4_u16j12svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za64_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -178,20 +106,12 @@ void test_svdot_multi_za64_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuin
 // Multi, multi (signed)
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za32_vg1x2_s16j11svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -200,28 +120,12 @@ void test_svdot_multi_za32_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint1
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za32_vg1x4_s16j11svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -230,20 +134,12 @@ void test_svdot_multi_za32_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint1
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_multi_za32_vg1x2_s8j10svint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svint8x2_t zm) __arm_streaming __arm_inout("za") {
@@ -252,28 +148,12 @@ void test_svdot_multi_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svint8x2
 
 // CHECK-LABEL: @test_svdot_multi_za32_vg1x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_multi_za32_vg1x4_s8j10svint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svint8x4_t zm) __arm_streaming __arm_inout("za") {
@@ -282,20 +162,12 @@ void test_svdot_multi_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svint8x4
 
 // CHECK-LABEL: @test_svdot_multi_za64_vg1x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za64_vg1x2_s16j11svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za64_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_streaming __arm_inout("za") {
@@ -304,28 +176,12 @@ void test_svdot_multi_za64_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint1
 
 // CHECK-LABEL: @test_svdot_multi_za64_vg1x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_multi_za64_vg1x4_s16j11svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_multi_za64_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm) __arm_streaming __arm_inout("za") {
@@ -337,16 +193,12 @@ void test_svdot_multi_za64_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint1
 // Multi, single (unsigned)
 // CHECK-LABEL: @test_svdot_single_za32_vg1x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za32_vg1x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -355,20 +207,12 @@ void test_svdot_single_za32_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svui
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za32_vg1x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -377,16 +221,12 @@ void test_svdot_single_za32_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svui
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_single_za32_vg1x2_u8j11svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -395,20 +235,12 @@ void test_svdot_single_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_single_za32_vg1x4_u8j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -417,16 +249,12 @@ void test_svdot_single_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint
 
 // CHECK-LABEL: @test_svdot_single_za64_vg1x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za64_vg1x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za64_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -435,20 +263,12 @@ void test_svdot_single_za64_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svui
 
 // CHECK-LABEL: @test_svdot_single_za64_vg1x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za64_vg1x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za64_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -460,16 +280,12 @@ void test_svdot_single_za64_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svui
 // Multi, single (signed)
 // CHECK-LABEL: @test_svdot_single_za32_vg1x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za32_vg1x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -478,20 +294,12 @@ void test_svdot_single_za32_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za32_vg1x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -500,16 +308,12 @@ void test_svdot_single_za32_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_single_za32_vg1x2_s8j10svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -518,20 +322,12 @@ void test_svdot_single_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_
 
 // CHECK-LABEL: @test_svdot_single_za32_vg1x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svdot_single_za32_vg1x4_s8j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -540,16 +336,12 @@ void test_svdot_single_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_
 
 // CHECK-LABEL: @test_svdot_single_za64_vg1x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za64_vg1x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za64_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -558,20 +350,12 @@ void test_svdot_single_za64_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint
 
 // CHECK-LABEL: @test_svdot_single_za64_vg1x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svdot_single_za64_vg1x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.single.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_single_za64_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -582,16 +366,12 @@ void test_svdot_single_za64_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint
 // Multi, indexed (unsigned)
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za32_vg1x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -600,20 +380,12 @@ void test_svdot_lane_za32_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za32_vg1x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -622,16 +394,12 @@ void test_svdot_lane_za32_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svdot_lane_za32_vg1x2_u8j11svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -640,20 +408,12 @@ void test_svdot_lane_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svdot_lane_za32_vg1x4_u8j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -662,16 +422,12 @@ void test_svdot_lane_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_
 
 // CHECK-LABEL: @test_svdot_lane_za64_vg1x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za64_vg1x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za64_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -680,20 +436,12 @@ void test_svdot_lane_za64_vg1x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint
 
 // CHECK-LABEL: @test_svdot_lane_za64_vg1x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za64_vg1x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.udot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za64_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -705,16 +453,12 @@ void test_svdot_lane_za64_vg1x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint
 // Multi, indexed (signed)
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za32_vg1x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -723,20 +467,12 @@ void test_svdot_lane_za32_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za32_vg1x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -745,16 +481,12 @@ void test_svdot_lane_za32_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svdot_lane_za32_vg1x2_s8j10svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -763,20 +495,12 @@ void test_svdot_lane_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t
 
 // CHECK-LABEL: @test_svdot_lane_za32_vg1x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svdot_lane_za32_vg1x4_s8j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -785,16 +509,12 @@ void test_svdot_lane_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t
 
 // CHECK-LABEL: @test_svdot_lane_za64_vg1x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za64_vg1x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za64_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -803,20 +523,12 @@ void test_svdot_lane_za64_vg1x2_s16(uint32_t slice_base, svint16x2_t zn, svint16
 
 // CHECK-LABEL: @test_svdot_lane_za64_vg1x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svdot_lane_za64_vg1x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svdot_lane_za64_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -828,20 +540,12 @@ void test_svdot_lane_za64_vg1x4_s16(uint32_t slice_base, svint16x4_t zn, svint16
 // Multi, multi (unsigned by signed)
 // CHECK-LABEL: @test_svusdot_multi_za32_vg1x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svusdot_multi_za32_vg1x2_u8j11svuint8x2_t10svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svusdot_multi_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8x2_t zm) __arm_streaming __arm_inout("za") {
@@ -850,28 +554,12 @@ void test_svusdot_multi_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svint
 
 // CHECK-LABEL: @test_svusdot_multi_za32_vg1x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svusdot_multi_za32_vg1x4_u8j11svuint8x4_t10svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svusdot_multi_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svint8x4_t zm) __arm_streaming __arm_inout("za") {
@@ -883,16 +571,12 @@ void test_svusdot_multi_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svint
 // Multi, single (unsigned by signed)
 // CHECK-LABEL: @test_svusdot_single_za32_vg1x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z33test_svusdot_single_za32_vg1x2_u8j11svuint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svusdot_single_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -901,20 +585,12 @@ void test_svusdot_single_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svin
 
 // CHECK-LABEL: @test_svusdot_single_za32_vg1x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z33test_svusdot_single_za32_vg1x4_u8j11svuint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svusdot_single_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -925,16 +601,12 @@ void test_svusdot_single_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svin
 // Multi, indexed (unsigned by signed)
 // CHECK-LABEL: @test_svusdot_lane_za32_vg1x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svusdot_lane_za32_vg1x2_u8j11svuint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svusdot_lane_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -943,20 +615,12 @@ void test_svusdot_lane_za32_vg1x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8
 
 // CHECK-LABEL: @test_svusdot_lane_za32_vg1x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svusdot_lane_za32_vg1x4_u8j11svuint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svusdot_lane_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -968,16 +632,12 @@ void test_svusdot_lane_za32_vg1x4_u8(uint32_t slice_base, svuint8x4_t zn, svint8
 // Multi, single (signed by unsigned)
 // CHECK-LABEL: @test_svsudot_single_za32_vg1x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z33test_svsudot_single_za32_vg1x2_s8j10svint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsudot_single_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -986,20 +646,12 @@ void test_svsudot_single_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svuin
 
 // CHECK-LABEL: @test_svsudot_single_za32_vg1x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z33test_svsudot_single_za32_vg1x4_s8j10svint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.single.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsudot_single_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -1010,20 +662,12 @@ void test_svsudot_single_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svuin
 // Multi, multi (signed by unsigned)
 // CHECK-LABEL: @test_svsudot_multi_za32_vg1x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svsudot_multi_za32_vg1x2_s8j10svint8x2_t11svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsudot_multi_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8x2_t zm) __arm_streaming __arm_inout("za") {
@@ -1032,28 +676,12 @@ void test_svsudot_multi_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svuint
 
 // CHECK-LABEL: @test_svsudot_multi_za32_vg1x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svsudot_multi_za32_vg1x4_s8j10svint8x4_t11svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usdot.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsudot_multi_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8x4_t zm) __arm_streaming __arm_inout("za") {
@@ -1064,16 +692,12 @@ void test_svsudot_multi_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svuint
 // Multi, indexed (signed by unsigned)
 // CHECK-LABEL: @test_svsudot_lane_za32_vg1x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svsudot_lane_za32_vg1x2_s8j10svint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsudot_lane_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -1082,20 +706,12 @@ void test_svsudot_lane_za32_vg1x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8
 
 // CHECK-LABEL: @test_svsudot_lane_za32_vg1x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svsudot_lane_za32_vg1x4_s8j10svint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sudot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsudot_lane_za32_vg1x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x2.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x2.c
index 4d340c6b8a6cd4..6dd55663d7d348 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x2.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x2.c
@@ -10,21 +10,27 @@
 
 // CHECK-LABEL: @test_svluti2_lane_zt_u8(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svluti2_lane_zt_u8u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svluti2_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_u8_x2(0, zn, 7);
@@ -33,21 +39,27 @@ svuint8x2_t test_svluti2_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0"
 
 // CHECK-LABEL: @test_svluti2_lane_zt_s8(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svluti2_lane_zt_s8u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svluti2_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_s8_x2(0, zn, 7);
@@ -55,21 +67,27 @@ svint8x2_t test_svluti2_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0")
 
 // CHECK-LABEL: @test_svluti2_lane_zt_u16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_u16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svluti2_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_u16_x2(0, zn, 7);
@@ -78,21 +96,27 @@ svuint16x2_t test_svluti2_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt
 
 // CHECK-LABEL: @test_svluti2_lane_zt_s16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_s16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svluti2_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_s16_x2(0, zn, 7);
@@ -100,21 +124,27 @@ svint16x2_t test_svluti2_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0
 
 // CHECK-LABEL: @test_svluti2_lane_zt_f16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_f16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svluti2_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_f16_x2(0, zn, 7);
@@ -122,21 +152,27 @@ svfloat16x2_t test_svluti2_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("z
 
 // CHECK-LABEL: @test_svluti2_lane_zt_bf16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svluti2_lane_zt_bf16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svluti2_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_bf16_x2(0, zn, 7);
@@ -144,21 +180,27 @@ svbfloat16x2_t test_svluti2_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svluti2_lane_zt_u32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_u32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svluti2_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_u32_x2(0, zn, 7);
@@ -166,21 +208,27 @@ svuint32x2_t test_svluti2_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt
 
 // CHECK-LABEL: @test_svluti2_lane_zt_s32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_s32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svluti2_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_s32_x2(0, zn, 7);
@@ -188,21 +236,27 @@ svint32x2_t test_svluti2_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0
 
 // CHECK-LABEL: @test_svluti2_lane_zt_f32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_f32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti2.lane.zt.x2.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 7)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svluti2_lane_zt_f32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_f32_x2(0, zn, 7);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x4.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x4.c
index 7fa74c1cb362a6..8650ec7f62dd83 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x4.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti2_lane_zt_x4.c
@@ -10,6 +10,7 @@
 
 // CHECK-LABEL: @test_svluti2_lane_zt_u8(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -19,10 +20,13 @@
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svluti2_lane_zt_u8u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -32,7 +36,9 @@
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svluti2_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_u8_x4(0, zn, 3);
@@ -41,6 +47,7 @@ svuint8x4_t test_svluti2_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0"
 
 // CHECK-LABEL: @test_svluti2_lane_zt_s8(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -50,10 +57,13 @@ svuint8x4_t test_svluti2_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0"
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svluti2_lane_zt_s8u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -63,7 +73,9 @@ svuint8x4_t test_svluti2_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0"
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svluti2_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_s8_x4(0, zn, 3);
@@ -71,6 +83,7 @@ svint8x4_t test_svluti2_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0")
 
 // CHECK-LABEL: @test_svluti2_lane_zt_u16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -80,10 +93,13 @@ svint8x4_t test_svluti2_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0")
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_u16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -93,7 +109,9 @@ svint8x4_t test_svluti2_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0")
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svluti2_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_u16_x4(0, zn, 3);
@@ -101,6 +119,7 @@ svuint16x4_t test_svluti2_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt
 
 // CHECK-LABEL: @test_svluti2_lane_zt_s16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -110,10 +129,13 @@ svuint16x4_t test_svluti2_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_s16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -123,7 +145,9 @@ svuint16x4_t test_svluti2_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svluti2_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_s16_x4(0, zn, 3);
@@ -131,6 +155,7 @@ svint16x4_t test_svluti2_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0
 
 // CHECK-LABEL: @test_svluti2_lane_zt_f16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -140,10 +165,13 @@ svint16x4_t test_svluti2_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_f16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -153,7 +181,9 @@ svint16x4_t test_svluti2_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svluti2_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_f16_x4(0, zn, 3);
@@ -161,6 +191,7 @@ svfloat16x4_t test_svluti2_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("z
 
 // CHECK-LABEL: @test_svluti2_lane_zt_bf16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -170,10 +201,13 @@ svfloat16x4_t test_svluti2_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("z
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svluti2_lane_zt_bf16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -183,7 +217,9 @@ svfloat16x4_t test_svluti2_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("z
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svluti2_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_bf16_x4(0, zn, 3);
@@ -191,6 +227,7 @@ svbfloat16x4_t test_svluti2_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svluti2_lane_zt_u32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -200,10 +237,13 @@ svbfloat16x4_t test_svluti2_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in(
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_u32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -213,7 +253,9 @@ svbfloat16x4_t test_svluti2_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in(
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svluti2_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_u32_x4(0, zn, 3);
@@ -221,6 +263,7 @@ svuint32x4_t test_svluti2_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt
 
 // CHECK-LABEL: @test_svluti2_lane_zt_s32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -230,10 +273,13 @@ svuint32x4_t test_svluti2_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_s32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -243,7 +289,9 @@ svuint32x4_t test_svluti2_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svluti2_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_s32_x4(0, zn, 3);
@@ -251,6 +299,7 @@ svint32x4_t test_svluti2_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0
 
 // CHECK-LABEL: @test_svluti2_lane_zt_f32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -260,10 +309,13 @@ svint32x4_t test_svluti2_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti2_lane_zt_f32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti2.lane.zt.x4.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -273,7 +325,9 @@ svint32x4_t test_svluti2_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svluti2_lane_zt_f32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti2_lane_zt_f32_x4(0, zn, 3);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x2.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x2.c
index f223a5dcdcefa7..f4f11c9fc5b143 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x2.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x2.c
@@ -10,21 +10,27 @@
 
 // CHECK-LABEL: @test_svluti4_lane_zt_u8(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svluti4_lane_zt_u8u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svluti4_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_u8_x2(0, zn, 3);
@@ -33,21 +39,27 @@ svuint8x2_t test_svluti4_lane_zt_u8(svuint8_t zn) __arm_streaming __arm_in("zt0"
 
 // CHECK-LABEL: @test_svluti4_lane_zt_s8(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svluti4_lane_zt_s8u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv16i8(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svluti4_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_s8_x2(0, zn, 3);
@@ -55,21 +67,27 @@ svint8x2_t test_svluti4_lane_zt_s8(svuint8_t zn) __arm_streaming __arm_in("zt0")
 
 // CHECK-LABEL: @test_svluti4_lane_zt_u16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti4_lane_zt_u16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svluti4_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_u16_x2(0, zn, 3);
@@ -78,21 +96,27 @@ svuint16x2_t test_svluti4_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt
 
 // CHECK-LABEL: @test_svluti4_lane_zt_s16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti4_lane_zt_s16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svluti4_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_s16_x2(0, zn, 3);
@@ -100,21 +124,27 @@ svint16x2_t test_svluti4_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0
 
 // CHECK-LABEL: @test_svluti4_lane_zt_f16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti4_lane_zt_f16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svluti4_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_f16_x2(0, zn, 3);
@@ -122,21 +152,27 @@ svfloat16x2_t test_svluti4_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("z
 
 // CHECK-LABEL: @test_svluti4_lane_zt_bf16(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svluti4_lane_zt_bf16u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svluti4_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_bf16_x2(0, zn, 3);
@@ -144,21 +180,27 @@ svbfloat16x2_t test_svluti4_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svluti4_lane_zt_u32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti4_lane_zt_u32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svluti4_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_u32_x2(0, zn, 3);
@@ -166,21 +208,27 @@ svuint32x2_t test_svluti4_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt
 
 // CHECK-LABEL: @test_svluti4_lane_zt_s32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti4_lane_zt_s32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svluti4_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_s32_x2(0, zn, 3);
@@ -188,21 +236,27 @@ svint32x2_t test_svluti4_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0
 
 // CHECK-LABEL: @test_svluti4_lane_zt_f32(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svluti4_lane_zt_f32u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti4.lane.zt.x2.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN:%.*]], i32 3)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svluti4_lane_zt_f32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_f32_x2(0, zn, 3);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x4.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x4.c
index 89e17e8d585882..16a74213262357 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x4.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_luti4_lane_zt_x4.c
@@ -8,9 +8,10 @@
 
 #include <arm_sme.h>
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svluti4_lane_zt_u16
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svluti4_lane_zt_u16
 // CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0:[0-9]+]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -20,11 +21,14 @@
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z24test_svluti4_lane_zt_u16u11__SVUint8_t
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z24test_svluti4_lane_zt_u16u11__SVUint8_t
 // CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0:[0-9]+]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -34,15 +38,18 @@
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svluti4_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_u16_x4(0, zn, 1);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x half> @test_svluti4_lane_zt_f16
+// CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @test_svluti4_lane_zt_f16
 // CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -52,11 +59,14 @@ svuint16x4_t test_svluti4_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x half> @_Z24test_svluti4_lane_zt_f16u11__SVUint8_t
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @_Z24test_svluti4_lane_zt_f16u11__SVUint8_t
 // CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8f16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -66,15 +76,18 @@ svuint16x4_t test_svluti4_lane_zt_u16(svuint8_t zn) __arm_streaming __arm_in("zt
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svluti4_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_f16_x4(0, zn, 1);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @test_svluti4_lane_zt_bf16
+// CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @test_svluti4_lane_zt_bf16
 // CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -84,11 +97,14 @@ svfloat16x4_t test_svluti4_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("z
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @_Z25test_svluti4_lane_zt_bf16u11__SVUint8_t
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @_Z25test_svluti4_lane_zt_bf16u11__SVUint8_t
 // CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8bf16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -98,15 +114,18 @@ svfloat16x4_t test_svluti4_lane_zt_f16(svuint8_t zn) __arm_streaming __arm_in("z
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svluti4_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_bf16_x4(0, zn, 1);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svluti4_lane_zt_s16
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svluti4_lane_zt_s16
 // CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -116,11 +135,14 @@ svbfloat16x4_t test_svluti4_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in(
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z24test_svluti4_lane_zt_s16u11__SVUint8_t
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z24test_svluti4_lane_zt_s16u11__SVUint8_t
 // CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv8i16(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -130,15 +152,18 @@ svbfloat16x4_t test_svluti4_lane_zt_bf16(svuint8_t zn) __arm_streaming __arm_in(
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svluti4_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_s16_x4(0, zn, 1);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svluti4_lane_zt_u32
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svluti4_lane_zt_u32
 // CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -148,11 +173,14 @@ svint16x4_t test_svluti4_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z24test_svluti4_lane_zt_u32u11__SVUint8_t
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z24test_svluti4_lane_zt_u32u11__SVUint8_t
 // CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -162,15 +190,18 @@ svint16x4_t test_svluti4_lane_zt_s16(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svluti4_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_u32_x4(0, zn, 1);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svluti4_lane_zt_s32
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svluti4_lane_zt_s32
 // CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -180,11 +211,14 @@ svuint32x4_t test_svluti4_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z24test_svluti4_lane_zt_s32u11__SVUint8_t
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z24test_svluti4_lane_zt_s32u11__SVUint8_t
 // CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv4i32(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -194,15 +228,18 @@ svuint32x4_t test_svluti4_lane_zt_u32(svuint8_t zn) __arm_streaming __arm_in("zt
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svluti4_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_s32_x4(0, zn, 1);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x float> @test_svluti4_lane_zt_f32
+// CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @test_svluti4_lane_zt_f32
 // CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -212,11 +249,14 @@ svint32x4_t test_svluti4_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x float> @_Z24test_svluti4_lane_zt_f32u11__SVUint8_t
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @_Z24test_svluti4_lane_zt_f32u11__SVUint8_t
 // CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.luti4.lane.zt.x4.nxv4f32(i32 0, <vscale x 16 x i8> [[ZN]], i32 1)
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -226,7 +266,9 @@ svint32x4_t test_svluti4_lane_zt_s32(svuint8_t zn) __arm_streaming __arm_in("zt0
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svluti4_lane_zt_f32(svuint8_t zn) __arm_streaming __arm_in("zt0") {
   return svluti4_lane_zt_f32_x4(0, zn, 1);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
index bd8d57e3523312..efc68c0b42334c 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
@@ -18,25 +18,27 @@
 
 // CHECK-LABEL: @test_svmax_single_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmax_single_s8_x210svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svmax_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s8_x2)(zdn, zm);
@@ -44,25 +46,27 @@ svint8x2_t test_svmax_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streaming
 
 // CHECK-LABEL: @test_svmax_single_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_s16_x211svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svmax_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s16_x2)(zdn, zm);
@@ -70,25 +74,27 @@ svint16x2_t test_svmax_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmax_single_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_s32_x211svint32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svmax_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s32_x2)(zdn, zm);
@@ -96,25 +102,27 @@ svint32x2_t test_svmax_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmax_single_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_s64_x211svint64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svmax_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s64_x2)(zdn, zm);
@@ -122,25 +130,27 @@ svint64x2_t test_svmax_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmax_single_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmax_single_u8_x211svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svmax_single_u8_x2(svuint8x2_t zdn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u8_x2)(zdn, zm);
@@ -148,25 +158,27 @@ svuint8x2_t test_svmax_single_u8_x2(svuint8x2_t zdn, svuint8_t zm) __arm_streami
 
 // CHECK-LABEL: @test_svmax_single_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_u16_x212svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svmax_single_u16_x2(svuint16x2_t zdn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u16_x2)(zdn, zm);
@@ -174,25 +186,27 @@ svuint16x2_t test_svmax_single_u16_x2(svuint16x2_t zdn, svuint16_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmax_single_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_u32_x212svuint32x2_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svmax_single_u32_x2(svuint32x2_t zdn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u32_x2)(zdn, zm);
@@ -200,25 +214,27 @@ svuint32x2_t test_svmax_single_u32_x2(svuint32x2_t zdn, svuint32_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmax_single_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_u64_x212svuint64x2_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svmax_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u64_x2)(zdn, zm);
@@ -226,25 +242,27 @@ svuint64x2_t test_svmax_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmax_single_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmax_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svmax_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_bf16_x2)(zdn, zm);
@@ -252,25 +270,27 @@ svbfloat16x2_t test_svmax_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __
 
 // CHECK-LABEL: @test_svmax_single_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_f16_x213svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svmax_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_f16_x2)(zdn, zm);
@@ -278,25 +298,27 @@ svfloat16x2_t test_svmax_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __arm_
 
 // CHECK-LABEL: @test_svmax_single_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_f32_x213svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svmax_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_f32_x2)(zdn, zm);
@@ -304,25 +326,27 @@ svfloat32x2_t test_svmax_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __arm_
 
 // CHECK-LABEL: @test_svmax_single_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_f64_x213svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svmax_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_f64_x2)(zdn, zm);
@@ -332,37 +356,35 @@ svfloat64x2_t test_svmax_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __arm_
 
 // CHECK-LABEL: @test_svmax_single_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmax_single_s8_x410svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svmax_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s8_x4)(zdn, zm);
@@ -370,37 +392,35 @@ svint8x4_t test_svmax_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streaming
 
 // CHECK-LABEL: @test_svmax_single_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_s16_x411svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svmax_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s16_x4)(zdn, zm);
@@ -408,37 +428,35 @@ svint16x4_t test_svmax_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmax_single_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_s32_x411svint32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svmax_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s32_x4)(zdn, zm);
@@ -446,37 +464,35 @@ svint32x4_t test_svmax_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmax_single_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_s64_x411svint64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svmax_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_s64_x4)(zdn, zm);
@@ -484,37 +500,35 @@ svint64x4_t test_svmax_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmax_single_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmax_single_u8_x411svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svmax_single_u8_x4(svuint8x4_t zdn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u8_x4)(zdn, zm);
@@ -522,37 +536,35 @@ svuint8x4_t test_svmax_single_u8_x4(svuint8x4_t zdn, svuint8_t zm) __arm_streami
 
 // CHECK-LABEL: @test_svmax_single_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_u16_x412svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svmax_single_u16_x4(svuint16x4_t zdn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u16_x4)(zdn, zm);
@@ -560,37 +572,35 @@ svuint16x4_t test_svmax_single_u16_x4(svuint16x4_t zdn, svuint16_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmax_single_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_u32_x412svuint32x4_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svmax_single_u32_x4(svuint32x4_t zdn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u32_x4)(zdn, zm);
@@ -598,37 +608,35 @@ svuint32x4_t test_svmax_single_u32_x4(svuint32x4_t zdn, svuint32_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmax_single_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_u64_x412svuint64x4_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svmax_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_u64_x4)(zdn, zm);
@@ -636,37 +644,35 @@ svuint64x4_t test_svmax_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmax_single_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmax_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svmax_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_bf16_x4)(zdn, zm);
@@ -674,37 +680,35 @@ svbfloat16x4_t test_svmax_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __
 
 // CHECK-LABEL: @test_svmax_single_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_f16_x413svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svmax_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_f16_x4)(zdn, zm);
@@ -712,37 +716,35 @@ svfloat16x4_t test_svmax_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __arm_
 
 // CHECK-LABEL: @test_svmax_single_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_f32_x413svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svmax_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_f32_x4)(zdn, zm);
@@ -750,37 +752,35 @@ svfloat32x4_t test_svmax_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __arm_
 
 // CHECK-LABEL: @test_svmax_single_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmax_single_f64_x413svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svmax_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_single_f64_x4)(zdn, zm);
@@ -790,29 +790,27 @@ svfloat64x4_t test_svmax_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __arm_
 
 // CHECK-LABEL: @test_svmax_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmax_s8_x210svint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svmax_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s8_x2)(zdn, zm);
@@ -820,29 +818,27 @@ svint8x2_t test_svmax_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_s16_x211svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svmax_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s16_x2)(zdn, zm);
@@ -850,29 +846,27 @@ svint16x2_t test_svmax_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_s32_x211svint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svmax_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s32_x2)(zdn, zm);
@@ -880,29 +874,27 @@ svint32x2_t test_svmax_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_s64_x211svint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svmax_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s64_x2)(zdn, zm);
@@ -910,29 +902,27 @@ svint64x2_t test_svmax_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmax_u8_x211svuint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svmax_u8_x2(svuint8x2_t zdn, svuint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u8_x2)(zdn, zm);
@@ -940,29 +930,27 @@ svuint8x2_t test_svmax_u8_x2(svuint8x2_t zdn, svuint8x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_u16_x212svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svmax_u16_x2(svuint16x2_t zdn, svuint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u16_x2)(zdn, zm);
@@ -970,29 +958,27 @@ svuint16x2_t test_svmax_u16_x2(svuint16x2_t zdn, svuint16x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmax_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_u32_x212svuint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svmax_u32_x2(svuint32x2_t zdn, svuint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u32_x2)(zdn, zm);
@@ -1000,29 +986,27 @@ svuint32x2_t test_svmax_u32_x2(svuint32x2_t zdn, svuint32x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmax_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_u64_x212svuint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svmax_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u64_x2)(zdn, zm);
@@ -1030,29 +1014,27 @@ svuint64x2_t test_svmax_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmax_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svmax_bf16_x214svbfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svmax_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_bf16_x2)(zdn, zm);
@@ -1060,29 +1042,27 @@ svbfloat16x2_t test_svmax_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_s
 
 // CHECK-LABEL: @test_svmax_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_f16_x213svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svmax_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_f16_x2)(zdn, zm);
@@ -1090,29 +1070,27 @@ svfloat16x2_t test_svmax_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmax_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_f32_x213svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svmax_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_f32_x2)(zdn, zm);
@@ -1120,29 +1098,27 @@ svfloat32x2_t test_svmax_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmax_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_f64_x213svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svmax_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_f64_x2)(zdn, zm);
@@ -1152,45 +1128,35 @@ svfloat64x2_t test_svmax_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmax_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmax_s8_x410svint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smax.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svmax_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s8_x4)(zdn, zm);
@@ -1198,45 +1164,35 @@ svint8x4_t test_svmax_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_s16_x411svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smax.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svmax_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s16_x4)(zdn, zm);
@@ -1244,45 +1200,35 @@ svint16x4_t test_svmax_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_s32_x411svint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smax.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svmax_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s32_x4)(zdn, zm);
@@ -1290,45 +1236,35 @@ svint32x4_t test_svmax_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_s64_x411svint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smax.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svmax_s64_x4(svint64x4_t zdn, svint64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_s64_x4)(zdn, zm);
@@ -1336,45 +1272,35 @@ svint64x4_t test_svmax_s64_x4(svint64x4_t zdn, svint64x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmax_u8_x411svuint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umax.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svmax_u8_x4(svuint8x4_t zdn, svuint8x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u8_x4)(zdn, zm);
@@ -1382,45 +1308,35 @@ svuint8x4_t test_svmax_u8_x4(svuint8x4_t zdn, svuint8x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmax_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_u16_x412svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umax.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svmax_u16_x4(svuint16x4_t zdn, svuint16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u16_x4)(zdn, zm);
@@ -1428,45 +1344,35 @@ svuint16x4_t test_svmax_u16_x4(svuint16x4_t zdn, svuint16x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmax_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_u32_x412svuint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umax.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svmax_u32_x4(svuint32x4_t zdn, svuint32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u32_x4)(zdn, zm);
@@ -1474,45 +1380,35 @@ svuint32x4_t test_svmax_u32_x4(svuint32x4_t zdn, svuint32x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmax_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_u64_x412svuint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umax.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svmax_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_u64_x4)(zdn, zm);
@@ -1520,45 +1416,35 @@ svuint64x4_t test_svmax_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmax_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svmax_bf16_x414svbfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svmax_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_bf16_x4)(zdn, zm);
@@ -1566,45 +1452,35 @@ svbfloat16x4_t test_svmax_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_s
 
 // CHECK-LABEL: @test_svmax_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_f16_x413svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmax.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svmax_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_f16_x4)(zdn, zm);
@@ -1612,45 +1488,35 @@ svfloat16x4_t test_svmax_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmax_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_f32_x413svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmax.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svmax_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_f32_x4)(zdn, zm);
@@ -1658,45 +1524,35 @@ svfloat32x4_t test_svmax_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmax_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmax_f64_x413svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmax.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svmax_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmax,_f64_x4)(zdn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c
index 07659932bef0a8..5d06895497cc7e 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c
@@ -19,25 +19,27 @@
 
 // CHECK-LABEL: @test_svmaxnm_single_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svmaxnm_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svmaxnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_bf16_x2)(zdn, zm);
@@ -45,25 +47,27 @@ svbfloat16x2_t test_svmaxnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm)
 
 // CHECK-LABEL: @test_svmaxnm_single_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_single_f16_x213svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svmaxnm_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_f16_x2)(zdn, zm);
@@ -71,25 +75,27 @@ svfloat16x2_t test_svmaxnm_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __ar
 
 // CHECK-LABEL: @test_svmaxnm_single_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_single_f32_x213svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svmaxnm_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_f32_x2)(zdn, zm);
@@ -97,25 +103,27 @@ svfloat32x2_t test_svmaxnm_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __ar
 
 // CHECK-LABEL: @test_svmaxnm_single_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_single_f64_x213svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svmaxnm_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_f64_x2)(zdn, zm);
@@ -125,37 +133,35 @@ svfloat64x2_t test_svmaxnm_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __ar
 
 // CHECK-LABEL: @test_svmaxnm_single_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svmaxnm_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svmaxnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_bf16_x4)(zdn, zm);
@@ -163,37 +169,35 @@ svbfloat16x4_t test_svmaxnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm)
 
 // CHECK-LABEL: @test_svmaxnm_single_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_single_f16_x413svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svmaxnm_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_f16_x4)(zdn, zm);
@@ -201,37 +205,35 @@ svfloat16x4_t test_svmaxnm_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __ar
 
 // CHECK-LABEL: @test_svmaxnm_single_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_single_f32_x413svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svmaxnm_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_f32_x4)(zdn, zm);
@@ -239,37 +241,35 @@ svfloat32x4_t test_svmaxnm_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __ar
 
 // CHECK-LABEL: @test_svmaxnm_single_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_single_f64_x413svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svmaxnm_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_single_f64_x4)(zdn, zm);
@@ -279,29 +279,27 @@ svfloat64x4_t test_svmaxnm_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __ar
 
 // CHECK-LABEL: @test_svmaxnm_multi_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_multi_bf16_x214svbfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svmaxnm_multi_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_bf16_x2)(zdn, zm);
@@ -309,29 +307,27 @@ svbfloat16x2_t test_svmaxnm_multi_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm)
 
 // CHECK-LABEL: @test_svmaxnm_multi_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmaxnm_multi_f16_x213svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svmaxnm_multi_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_f16_x2)(zdn, zm);
@@ -339,29 +335,27 @@ svfloat16x2_t test_svmaxnm_multi_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __a
 
 // CHECK-LABEL: @test_svmaxnm_multi_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmaxnm_multi_f32_x213svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svmaxnm_multi_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_f32_x2)(zdn, zm);
@@ -369,29 +363,27 @@ svfloat32x2_t test_svmaxnm_multi_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __a
 
 // CHECK-LABEL: @test_svmaxnm_multi_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmaxnm_multi_f64_x213svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svmaxnm_multi_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_f64_x2)(zdn, zm);
@@ -401,45 +393,35 @@ svfloat64x2_t test_svmaxnm_multi_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __a
 
 // CHECK-LABEL: @test_svmaxnm_multi_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svmaxnm_multi_bf16_x414svbfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svmaxnm_multi_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_bf16_x4)(zdn, zm);
@@ -447,45 +429,35 @@ svbfloat16x4_t test_svmaxnm_multi_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm)
 
 // CHECK-LABEL: @test_svmaxnm_multi_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmaxnm_multi_f16_x413svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmaxnm.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svmaxnm_multi_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_f16_x4)(zdn, zm);
@@ -493,45 +465,35 @@ svfloat16x4_t test_svmaxnm_multi_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __a
 
 // CHECK-LABEL: @test_svmaxnm_multi_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmaxnm_multi_f32_x413svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmaxnm.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svmaxnm_multi_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_f32_x4)(zdn, zm);
@@ -539,45 +501,35 @@ svfloat32x4_t test_svmaxnm_multi_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __a
 
 // CHECK-LABEL: @test_svmaxnm_multi_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmaxnm_multi_f64_x413svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmaxnm.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svmaxnm_multi_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmaxnm,_f64_x4)(zdn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c
index fe7b74c0052476..2fa7feeee404e6 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c
@@ -18,25 +18,27 @@
 
 // CHECK-LABEL: @test_svmin_single_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmin_single_s8_x210svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svmin_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s8_x2)(zdn, zm);
@@ -44,25 +46,27 @@ svint8x2_t test_svmin_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streaming
 
 // CHECK-LABEL: @test_svmin_single_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_s16_x211svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svmin_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s16_x2)(zdn, zm);
@@ -70,25 +74,27 @@ svint16x2_t test_svmin_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmin_single_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_s32_x211svint32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svmin_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s32_x2)(zdn, zm);
@@ -96,25 +102,27 @@ svint32x2_t test_svmin_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmin_single_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_s64_x211svint64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svmin_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s64_x2)(zdn, zm);
@@ -122,25 +130,27 @@ svint64x2_t test_svmin_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmin_single_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmin_single_u8_x211svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svmin_single_u8_x2(svuint8x2_t zdn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u8_x2)(zdn, zm);
@@ -148,25 +158,27 @@ svuint8x2_t test_svmin_single_u8_x2(svuint8x2_t zdn, svuint8_t zm) __arm_streami
 
 // CHECK-LABEL: @test_svmin_single_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_u16_x212svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svmin_single_u16_x2(svuint16x2_t zdn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u16_x2)(zdn, zm);
@@ -174,25 +186,27 @@ svuint16x2_t test_svmin_single_u16_x2(svuint16x2_t zdn, svuint16_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmin_single_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_u32_x212svuint32x2_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svmin_single_u32_x2(svuint32x2_t zdn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u32_x2)(zdn, zm);
@@ -200,25 +214,27 @@ svuint32x2_t test_svmin_single_u32_x2(svuint32x2_t zdn, svuint32_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmin_single_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_u64_x212svuint64x2_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svmin_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u64_x2)(zdn, zm);
@@ -226,25 +242,27 @@ svuint64x2_t test_svmin_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmin_single_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmin_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svmin_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_bf16_x2)(zdn, zm);
@@ -252,25 +270,27 @@ svbfloat16x2_t test_svmin_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __
 
 // CHECK-LABEL: @test_svmin_single_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_f16_x213svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svmin_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_f16_x2)(zdn, zm);
@@ -278,25 +298,27 @@ svfloat16x2_t test_svmin_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __arm_
 
 // CHECK-LABEL: @test_svmin_single_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_f32_x213svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svmin_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_f32_x2)(zdn, zm);
@@ -304,25 +326,27 @@ svfloat32x2_t test_svmin_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __arm_
 
 // CHECK-LABEL: @test_svmin_single_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_f64_x213svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svmin_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_f64_x2)(zdn, zm);
@@ -332,37 +356,35 @@ svfloat64x2_t test_svmin_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __arm_
 
 // CHECK-LABEL: @test_svmin_single_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmin_single_s8_x410svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svmin_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s8_x4)(zdn, zm);
@@ -370,37 +392,35 @@ svint8x4_t test_svmin_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streaming
 
 // CHECK-LABEL: @test_svmin_single_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_s16_x411svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svmin_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s16_x4)(zdn, zm);
@@ -408,37 +428,35 @@ svint16x4_t test_svmin_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmin_single_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_s32_x411svint32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svmin_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s32_x4)(zdn, zm);
@@ -446,37 +464,35 @@ svint32x4_t test_svmin_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmin_single_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_s64_x411svint64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svmin_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_s64_x4)(zdn, zm);
@@ -484,37 +500,35 @@ svint64x4_t test_svmin_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svmin_single_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svmin_single_u8_x411svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svmin_single_u8_x4(svuint8x4_t zdn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u8_x4)(zdn, zm);
@@ -522,37 +536,35 @@ svuint8x4_t test_svmin_single_u8_x4(svuint8x4_t zdn, svuint8_t zm) __arm_streami
 
 // CHECK-LABEL: @test_svmin_single_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_u16_x412svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svmin_single_u16_x4(svuint16x4_t zdn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u16_x4)(zdn, zm);
@@ -560,37 +572,35 @@ svuint16x4_t test_svmin_single_u16_x4(svuint16x4_t zdn, svuint16_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmin_single_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_u32_x412svuint32x4_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svmin_single_u32_x4(svuint32x4_t zdn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u32_x4)(zdn, zm);
@@ -598,37 +608,35 @@ svuint32x4_t test_svmin_single_u32_x4(svuint32x4_t zdn, svuint32_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmin_single_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_u64_x412svuint64x4_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svmin_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_u64_x4)(zdn, zm);
@@ -636,37 +644,35 @@ svuint64x4_t test_svmin_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_str
 
 // CHECK-LABEL: @test_svmin_single_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svmin_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svmin_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_bf16_x4)(zdn, zm);
@@ -674,37 +680,35 @@ svbfloat16x4_t test_svmin_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __
 
 // CHECK-LABEL: @test_svmin_single_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_f16_x413svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svmin_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_f16_x4)(zdn, zm);
@@ -712,37 +716,35 @@ svfloat16x4_t test_svmin_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __arm_
 
 // CHECK-LABEL: @test_svmin_single_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_f32_x413svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svmin_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_f32_x4)(zdn, zm);
@@ -750,37 +752,35 @@ svfloat32x4_t test_svmin_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __arm_
 
 // CHECK-LABEL: @test_svmin_single_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svmin_single_f64_x413svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svmin_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_single_f64_x4)(zdn, zm);
@@ -790,29 +790,27 @@ svfloat64x4_t test_svmin_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __arm_
 
 // CHECK-LABEL: @test_svmin_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmin_s8_x210svint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svmin_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s8_x2)(zdn, zm);
@@ -820,29 +818,27 @@ svint8x2_t test_svmin_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_s16_x211svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svmin_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s16_x2)(zdn, zm);
@@ -850,29 +846,27 @@ svint16x2_t test_svmin_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_s32_x211svint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svmin_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s32_x2)(zdn, zm);
@@ -880,29 +874,27 @@ svint32x2_t test_svmin_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_s64_x211svint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svmin_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s64_x2)(zdn, zm);
@@ -910,29 +902,27 @@ svint64x2_t test_svmin_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmin_u8_x211svuint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svmin_u8_x2(svuint8x2_t zdn, svuint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u8_x2)(zdn, zm);
@@ -940,29 +930,27 @@ svuint8x2_t test_svmin_u8_x2(svuint8x2_t zdn, svuint8x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_u16_x212svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svmin_u16_x2(svuint16x2_t zdn, svuint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u16_x2)(zdn, zm);
@@ -970,29 +958,27 @@ svuint16x2_t test_svmin_u16_x2(svuint16x2_t zdn, svuint16x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmin_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_u32_x212svuint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svmin_u32_x2(svuint32x2_t zdn, svuint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u32_x2)(zdn, zm);
@@ -1000,29 +986,27 @@ svuint32x2_t test_svmin_u32_x2(svuint32x2_t zdn, svuint32x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmin_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_u64_x212svuint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svmin_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u64_x2)(zdn, zm);
@@ -1030,29 +1014,27 @@ svuint64x2_t test_svmin_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmin_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svmin_bf16_x214svbfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svmin_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_bf16_x2)(zdn, zm);
@@ -1060,29 +1042,27 @@ svbfloat16x2_t test_svmin_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_s
 
 // CHECK-LABEL: @test_svmin_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_f16_x213svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svmin_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_f16_x2)(zdn, zm);
@@ -1090,29 +1070,27 @@ svfloat16x2_t test_svmin_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmin_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_f32_x213svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svmin_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_f32_x2)(zdn, zm);
@@ -1120,29 +1098,27 @@ svfloat32x2_t test_svmin_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmin_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_f64_x213svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svmin_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_f64_x2)(zdn, zm);
@@ -1152,45 +1128,35 @@ svfloat64x2_t test_svmin_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmin_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmin_s8_x410svint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.smin.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svmin_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s8_x4)(zdn, zm);
@@ -1198,45 +1164,35 @@ svint8x4_t test_svmin_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_s16_x411svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.smin.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svmin_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s16_x4)(zdn, zm);
@@ -1244,45 +1200,35 @@ svint16x4_t test_svmin_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_s32_x411svint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.smin.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svmin_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s32_x4)(zdn, zm);
@@ -1290,45 +1236,35 @@ svint32x4_t test_svmin_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_s64_x411svint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.smin.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svmin_s64_x4(svint64x4_t zdn, svint64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_s64_x4)(zdn, zm);
@@ -1336,45 +1272,35 @@ svint64x4_t test_svmin_s64_x4(svint64x4_t zdn, svint64x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svmin_u8_x411svuint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.umin.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svmin_u8_x4(svuint8x4_t zdn, svuint8x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u8_x4)(zdn, zm);
@@ -1382,45 +1308,35 @@ svuint8x4_t test_svmin_u8_x4(svuint8x4_t zdn, svuint8x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svmin_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_u16_x412svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.umin.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svmin_u16_x4(svuint16x4_t zdn, svuint16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u16_x4)(zdn, zm);
@@ -1428,45 +1344,35 @@ svuint16x4_t test_svmin_u16_x4(svuint16x4_t zdn, svuint16x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmin_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_u32_x412svuint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.umin.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svmin_u32_x4(svuint32x4_t zdn, svuint32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u32_x4)(zdn, zm);
@@ -1474,45 +1380,35 @@ svuint32x4_t test_svmin_u32_x4(svuint32x4_t zdn, svuint32x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmin_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_u64_x412svuint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.umin.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svmin_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_u64_x4)(zdn, zm);
@@ -1520,45 +1416,35 @@ svuint64x4_t test_svmin_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svmin_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svmin_bf16_x414svbfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svmin_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_bf16_x4)(zdn, zm);
@@ -1566,45 +1452,35 @@ svbfloat16x4_t test_svmin_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_s
 
 // CHECK-LABEL: @test_svmin_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_f16_x413svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fmin.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svmin_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_f16_x4)(zdn, zm);
@@ -1612,45 +1488,35 @@ svfloat16x4_t test_svmin_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmin_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_f32_x413svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fmin.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svmin_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_f32_x4)(zdn, zm);
@@ -1658,45 +1524,35 @@ svfloat32x4_t test_svmin_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svmin_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svmin_f64_x413svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fmin.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svmin_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svmin,_f64_x4)(zdn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c
index 3b221c030eddf9..71b8914b816cad 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c
@@ -19,25 +19,27 @@
 
 // CHECK-LABEL: @test_svminnm_single_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svminnm_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svminnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_bf16_x2)(zdn, zm);
@@ -45,25 +47,27 @@ svbfloat16x2_t test_svminnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm)
 
 // CHECK-LABEL: @test_svminnm_single_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_single_f16_x213svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svminnm_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_f16_x2)(zdn, zm);
@@ -71,25 +75,27 @@ svfloat16x2_t test_svminnm_single_f16_x2(svfloat16x2_t zdn, svfloat16_t zm) __ar
 
 // CHECK-LABEL: @test_svminnm_single_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_single_f32_x213svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svminnm_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_f32_x2)(zdn, zm);
@@ -97,25 +103,27 @@ svfloat32x2_t test_svminnm_single_f32_x2(svfloat32x2_t zdn, svfloat32_t zm) __ar
 
 // CHECK-LABEL: @test_svminnm_single_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_single_f64_x213svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svminnm_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_f64_x2)(zdn, zm);
@@ -125,37 +133,35 @@ svfloat64x2_t test_svminnm_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __ar
 
 // CHECK-LABEL: @test_svminnm_single_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svminnm_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svminnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_bf16_x4)(zdn, zm);
@@ -163,37 +169,35 @@ svbfloat16x4_t test_svminnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm)
 
 // CHECK-LABEL: @test_svminnm_single_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_single_f16_x413svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.single.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svminnm_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_f16_x4)(zdn, zm);
@@ -201,37 +205,35 @@ svfloat16x4_t test_svminnm_single_f16_x4(svfloat16x4_t zdn, svfloat16_t zm) __ar
 
 // CHECK-LABEL: @test_svminnm_single_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_single_f32_x413svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.single.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svminnm_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_f32_x4)(zdn, zm);
@@ -239,37 +241,35 @@ svfloat32x4_t test_svminnm_single_f32_x4(svfloat32x4_t zdn, svfloat32_t zm) __ar
 
 // CHECK-LABEL: @test_svminnm_single_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_single_f64_x413svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.single.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svminnm_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_single_f64_x4)(zdn, zm);
@@ -279,29 +279,27 @@ svfloat64x4_t test_svminnm_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __ar
 
 // CHECK-LABEL: @test_svminnm_multi_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_multi_bf16_x214svbfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svminnm_multi_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_bf16_x2)(zdn, zm);
@@ -309,29 +307,27 @@ svbfloat16x2_t test_svminnm_multi_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm)
 
 // CHECK-LABEL: @test_svminnm_multi_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svminnm_multi_f16_x213svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x2.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svminnm_multi_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_f16_x2)(zdn, zm);
@@ -339,29 +335,27 @@ svfloat16x2_t test_svminnm_multi_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __a
 
 // CHECK-LABEL: @test_svminnm_multi_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svminnm_multi_f32_x213svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x2.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svminnm_multi_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_f32_x2)(zdn, zm);
@@ -369,29 +363,27 @@ svfloat32x2_t test_svminnm_multi_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __a
 
 // CHECK-LABEL: @test_svminnm_multi_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svminnm_multi_f64_x213svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x2.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svminnm_multi_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_f64_x2)(zdn, zm);
@@ -401,45 +393,35 @@ svfloat64x2_t test_svminnm_multi_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __a
 
 // CHECK-LABEL: @test_svminnm_multi_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svminnm_multi_bf16_x414svbfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZDN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZDN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svminnm_multi_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_bf16_x4)(zdn, zm);
@@ -447,45 +429,35 @@ svbfloat16x4_t test_svminnm_multi_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm)
 
 // CHECK-LABEL: @test_svminnm_multi_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svminnm_multi_f16_x413svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.fminnm.x4.nxv8f16(<vscale x 8 x half> [[ZDN_COERCE0:%.*]], <vscale x 8 x half> [[ZDN_COERCE1:%.*]], <vscale x 8 x half> [[ZDN_COERCE2:%.*]], <vscale x 8 x half> [[ZDN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svminnm_multi_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_f16_x4)(zdn, zm);
@@ -493,45 +465,35 @@ svfloat16x4_t test_svminnm_multi_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __a
 
 // CHECK-LABEL: @test_svminnm_multi_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svminnm_multi_f32_x413svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fminnm.x4.nxv4f32(<vscale x 4 x float> [[ZDN_COERCE0:%.*]], <vscale x 4 x float> [[ZDN_COERCE1:%.*]], <vscale x 4 x float> [[ZDN_COERCE2:%.*]], <vscale x 4 x float> [[ZDN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svminnm_multi_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_f32_x4)(zdn, zm);
@@ -539,45 +501,35 @@ svfloat32x4_t test_svminnm_multi_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __a
 
 // CHECK-LABEL: @test_svminnm_multi_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svminnm_multi_f64_x413svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.fminnm.x4.nxv2f64(<vscale x 2 x double> [[ZDN_COERCE0:%.*]], <vscale x 2 x double> [[ZDN_COERCE1:%.*]], <vscale x 2 x double> [[ZDN_COERCE2:%.*]], <vscale x 2 x double> [[ZDN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svminnm_multi_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svminnm,_f64_x4)(zdn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mla.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mla.c
index cd8adf4a81ce71..74859c0a23bbba 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mla.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mla.c
@@ -19,20 +19,12 @@
 // Multi, multi
 // CHECK-LABEL: @test_svmla2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla2_f32j13svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32x2_t zm) __arm_streaming __arm_inout("za") {
@@ -41,28 +33,12 @@ void test_svmla2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32x2_t zm) __
 
 // CHECK-LABEL: @test_svmla4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla4_f32j13svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32x4_t zm) __arm_streaming __arm_inout("za") {
@@ -73,16 +49,12 @@ void test_svmla4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32x4_t zm) __
 // Multi, single
 // CHECK-LABEL: @test_svmla_single2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single2_f32j13svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -91,20 +63,12 @@ void test_svmla_single2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t z
 
 // CHECK-LABEL: @test_svmla_single4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single4_f32j13svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -115,16 +79,12 @@ void test_svmla_single4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t z
 // Multi, indexed
 // CHECK-LABEL: @test_svmla_lane2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane2_f32j13svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -133,20 +93,12 @@ void test_svmla_lane2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t zm)
 
 // CHECK-LABEL: @test_svmla_lane4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane4_f32j13svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -157,20 +109,12 @@ void test_svmla_lane4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t zm)
 // Multi, multi
 // CHECK-LABEL: @test_svmla2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla2_f64j13svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64x2_t zm) __arm_streaming __arm_inout("za") {
@@ -179,28 +123,12 @@ void test_svmla2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64x2_t zm) __
 
 // CHECK-LABEL: @test_svmla4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla4_f64j13svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64x4_t zm) __arm_streaming __arm_inout("za") {
@@ -211,16 +139,12 @@ void test_svmla4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64x4_t zm) __
 // Multi, single
 // CHECK-LABEL: @test_svmla_single2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single2_f64j13svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {
@@ -229,20 +153,12 @@ void test_svmla_single2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t z
 
 // CHECK-LABEL: @test_svmla_single4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single4_f64j13svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {
@@ -253,16 +169,12 @@ void test_svmla_single4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64_t z
 // Multi, indexed
 // CHECK-LABEL: @test_svmla_lane2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane2_f64j13svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {
@@ -271,20 +183,12 @@ void test_svmla_lane2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t zm)
 
 // CHECK-LABEL: @test_svmla_lane4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane4_f64j13svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmla.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlal.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlal.c
index c0444c99b63470..9d8b1351debc2a 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlal.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlal.c
@@ -19,20 +19,12 @@
 // Multi, multi
 // CHECK-LABEL: @test_svmla2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla2_f16j13svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16x2_t zm) __arm_streaming __arm_inout("za")
@@ -42,20 +34,12 @@ void test_svmla2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16x2_t zm) __
 
 // CHECK-LABEL: @test_svmla2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z16test_svmla2_bf16j14svbfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16x2_t zm) __arm_streaming __arm_inout("za")
@@ -65,20 +49,12 @@ void test_svmla2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16x2_t zm)
 
 // CHECK-LABEL: @test_svmla2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla2_u16j12svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -88,20 +64,12 @@ void test_svmla2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __ar
 
 // CHECK-LABEL: @test_svmla2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla2_s16j11svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -111,28 +79,12 @@ void test_svmla2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_
 
 // CHECK-LABEL: @test_svmla4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla4_f16j13svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16x4_t zm) __arm_streaming __arm_inout("za")
@@ -142,28 +94,12 @@ void test_svmla4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16x4_t zm) __
 
 // CHECK-LABEL: @test_svmla4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z16test_svmla4_bf16j14svbfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16x4_t zm) __arm_streaming __arm_inout("za")
@@ -173,28 +109,12 @@ void test_svmla4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16x4_t zm)
 
 // CHECK-LABEL: @test_svmla4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla4_u16j12svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -204,28 +124,12 @@ void test_svmla4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __ar
 
 // CHECK-LABEL: @test_svmla4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmla4_s16j11svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -297,16 +201,12 @@ void test_svmla_single1_s16(uint32_t slice_base, svint16_t zn, svint16_t zm) __a
 
 // CHECK-LABEL: @test_svmla_single2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single2_f16j13svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -316,16 +216,12 @@ void test_svmla_single2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t z
 
 // CHECK-LABEL: @test_svmla_single2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmla_single2_bf16j14svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -335,16 +231,12 @@ void test_svmla_single2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_
 
 // CHECK-LABEL: @test_svmla_single2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -354,16 +246,12 @@ void test_svmla_single2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_svmla_single2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -373,20 +261,12 @@ void test_svmla_single2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) _
 
 // CHECK-LABEL: @test_svmla_single4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single4_f16j13svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -396,20 +276,12 @@ void test_svmla_single4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t z
 
 // CHECK-LABEL: @test_svmla_single4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmla_single4_bf16j14svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -419,20 +291,12 @@ void test_svmla_single4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_
 
 // CHECK-LABEL: @test_svmla_single4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -442,20 +306,12 @@ void test_svmla_single4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_svmla_single4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmla_single4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -529,16 +385,12 @@ void test_svmla_lane1_s16(uint32_t slice_base, svint16_t zn, svint16_t zm) __arm
 
 // CHECK-LABEL: @test_svmla_lane2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane2_f16j13svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -548,16 +400,12 @@ void test_svmla_lane2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm)
 
 // CHECK-LABEL: @test_svmla_lane2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_svmla_lane2_bf16j14svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -567,16 +415,12 @@ void test_svmla_lane2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t
 
 // CHECK-LABEL: @test_svmla_lane2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -586,16 +430,12 @@ void test_svmla_lane2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) _
 
 // CHECK-LABEL: @test_svmla_lane2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -605,20 +445,12 @@ void test_svmla_lane2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __a
 
 // CHECK-LABEL: @test_svmla_lane4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane4_f16j13svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -628,20 +460,12 @@ void test_svmla_lane4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm)
 
 // CHECK-LABEL: @test_svmla_lane4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_svmla_lane4_bf16j14svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlal.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -651,20 +475,12 @@ void test_svmla_lane4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t
 
 // CHECK-LABEL: @test_svmla_lane4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -674,20 +490,12 @@ void test_svmla_lane4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) _
 
 // CHECK-LABEL: @test_svmla_lane4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmla_lane4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlal.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_lane4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlall.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlall.c
index 8e4d92fce6c6b7..4efc226c10e685 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlall.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlall.c
@@ -185,16 +185,12 @@ void test_usmlall_single_x1_u8(uint32_t slice_base, svuint8_t zn, svint8_t zm) _
 
 // CHECK-LABEL: @test_svmla_single_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmla_single_x2_s8j10svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -204,16 +200,12 @@ void test_svmla_single_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __
 
 // CHECK-LABEL: @test_svmla_single_x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmla_single_x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -223,16 +215,12 @@ void test_svmla_single_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm)
 
 // CHECK-LABEL: @test_svmla_single_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmla_single_x2_u8j11svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -242,16 +230,12 @@ void test_svmla_single_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm)
 
 // CHECK-LABEL: @test_svmla_single_x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmla_single_x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -263,16 +247,12 @@ void test_svmla_single_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t z
 
 // CHECK-LABEL: @test_svmls_single_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmls_single_x2_s8j10svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -282,16 +262,12 @@ void test_svmls_single_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __
 
 // CHECK-LABEL: @test_svmls_single_x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmls_single_x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -301,16 +277,12 @@ void test_svmls_single_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm)
 
 // CHECK-LABEL: @test_svmls_single_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmls_single_x2_u8j11svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -320,16 +292,12 @@ void test_svmls_single_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm)
 
 // CHECK-LABEL: @test_svmls_single_x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmls_single_x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -341,16 +309,12 @@ void test_svmls_single_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t z
 
 // CHECK-LABEL: @test_svsumla_single_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsumla_single_x2_s8j10svint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsumla_single_x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -362,16 +326,12 @@ void test_svsumla_single_x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8_t zm)
 
 // CHECK-LABEL: @test_usmlall_single_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_usmlall_single_x2_u8j11svuint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_usmlall_single_x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -387,20 +347,12 @@ void test_usmlall_single_x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8_t zm)
 
 // CHECK-LABEL: @test_svmla_single_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmla_single_x4_s8j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -410,20 +362,12 @@ void test_svmla_single_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __
 
 // CHECK-LABEL: @test_svmla_single_x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmla_single_x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -433,20 +377,12 @@ void test_svmla_single_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm)
 
 // CHECK-LABEL: @test_svmla_single_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmla_single_x4_u8j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -456,20 +392,12 @@ void test_svmla_single_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm)
 
 // CHECK-LABEL: @test_svmla_single_x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmla_single_x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmla_single_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -481,20 +409,12 @@ void test_svmla_single_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t z
 
 // CHECK-LABEL: @test_svmls_single_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmls_single_x4_s8j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -504,20 +424,12 @@ void test_svmls_single_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __
 
 // CHECK-LABEL: @test_svmls_single_x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmls_single_x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -527,20 +439,12 @@ void test_svmls_single_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm)
 
 // CHECK-LABEL: @test_svmls_single_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmls_single_x4_u8j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -550,20 +454,12 @@ void test_svmls_single_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm)
 
 // CHECK-LABEL: @test_svmls_single_x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z24test_svmls_single_x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.single.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -575,20 +471,12 @@ void test_svmls_single_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t z
 
 // CHECK-LABEL: @test_svsumla_single_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsumla_single_x4_s8j10svint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsumla_single_x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -600,20 +488,12 @@ void test_svsumla_single_x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8_t zm)
 
 // CHECK-LABEL: @test_usmlall_single_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_usmlall_single_x4_u8j11svuint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.single.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_usmlall_single_x4_u8(uint32_t slice_base, svuint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -629,20 +509,12 @@ void test_usmlall_single_x4_u8(uint32_t slice_base, svuint8x4_t zn, svint8_t zm)
 
 // CHECK-LABEL: @test_mlal_multi_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlal_multi_x2_s8j10svint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8x2_t zm) __arm_streaming __arm_inout("za")
@@ -652,20 +524,12 @@ void test_mlal_multi_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8x2_t zm) __
 
 // CHECK-LABEL: @test_mlal_multi_x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlal_multi_x2_s16j11svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -675,20 +539,12 @@ void test_mlal_multi_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm)
 
 // CHECK-LABEL: @test_mlal_multi_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlal_multi_x2_u8j11svuint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8x2_t zm) __arm_streaming __arm_inout("za")
@@ -698,20 +554,12 @@ void test_mlal_multi_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8x2_t zm)
 
 // CHECK-LABEL: @test_mlal_multi_x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlal_multi_x2_u16j12svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -723,20 +571,12 @@ void test_mlal_multi_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t z
 
 // CHECK-LABEL: @test_mlsl_multi_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlsl_multi_x2_s8j10svint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8x2_t zm) __arm_streaming __arm_inout("za")
@@ -746,20 +586,12 @@ void test_mlsl_multi_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8x2_t zm) __
 
 // CHECK-LABEL: @test_mlsl_multi_x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlsl_multi_x2_s16j11svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -769,20 +601,12 @@ void test_mlsl_multi_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm)
 
 // CHECK-LABEL: @test_mlsl_multi_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlsl_multi_x2_u8j11svuint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8x2_t zm) __arm_streaming __arm_inout("za")
@@ -792,20 +616,12 @@ void test_mlsl_multi_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8x2_t zm)
 
 // CHECK-LABEL: @test_mlsl_multi_x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlsl_multi_x2_u16j12svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -817,20 +633,12 @@ void test_mlsl_multi_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t z
 
 // CHECK-LABEL: @test_sumlal_multi_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_sumlal_multi_x2_s8j10svint8x2_t11svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_sumlal_multi_x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8x2_t zm) __arm_streaming __arm_inout("za")
@@ -842,20 +650,12 @@ void test_sumlal_multi_x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8x2_t zm)
 
 // CHECK-LABEL: @test_usmlal_multi_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_usmlal_multi_x2_u8j11svuint8x2_t10svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_usmlal_multi_x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8x2_t zm) __arm_streaming __arm_inout("za")
@@ -871,28 +671,12 @@ void test_usmlal_multi_x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8x2_t zm)
 
 // CHECK-LABEL: @test_mlal_multi_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlal_multi_x4_s8j10svint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8x4_t zm) __arm_streaming __arm_inout("za")
@@ -902,28 +686,12 @@ void test_mlal_multi_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8x4_t zm) __
 
 // CHECK-LABEL: @test_mlal_multi_x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlal_multi_x4_s16j11svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -933,28 +701,12 @@ void test_mlal_multi_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm)
 
 // CHECK-LABEL: @test_mlal_multi_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlal_multi_x4_u8j11svuint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8x4_t zm) __arm_streaming __arm_inout("za")
@@ -964,28 +716,12 @@ void test_mlal_multi_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8x4_t zm)
 
 // CHECK-LABEL: @test_mlal_multi_x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlal_multi_x4_u16j12svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlal_multi_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -997,28 +733,12 @@ void test_mlal_multi_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t z
 
 // CHECK-LABEL: @test_mlsl_multi_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlsl_multi_x4_s8j10svint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8x4_t zm) __arm_streaming __arm_inout("za")
@@ -1028,28 +748,12 @@ void test_mlsl_multi_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8x4_t zm) __
 
 // CHECK-LABEL: @test_mlsl_multi_x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlsl_multi_x4_s16j11svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -1059,28 +763,12 @@ void test_mlsl_multi_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm)
 
 // CHECK-LABEL: @test_mlsl_multi_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_mlsl_multi_x4_u8j11svuint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8x4_t zm) __arm_streaming __arm_inout("za")
@@ -1090,28 +778,12 @@ void test_mlsl_multi_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8x4_t zm)
 
 // CHECK-LABEL: @test_mlsl_multi_x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_mlsl_multi_x4_u16j12svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_mlsl_multi_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -1123,28 +795,12 @@ void test_mlsl_multi_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t z
 
 // CHECK-LABEL: @test_sumlal_multi_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_sumlal_multi_x4_s8j10svint8x4_t11svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_sumlal_multi_x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8x4_t zm) __arm_streaming __arm_inout("za")
@@ -1156,28 +812,12 @@ void test_sumlal_multi_x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8x4_t zm)
 
 // CHECK-LABEL: @test_usmlal_multi_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_usmlal_multi_x4_u8j11svuint8x4_t10svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_usmlal_multi_x4_u8(uint32_t slice_base, svuint8x4_t zn, svint8x4_t zm) __arm_streaming __arm_inout("za")
@@ -1355,16 +995,12 @@ void test_usmlall_lane_x1_u8(uint32_t slice_base, svuint8_t zn, svint8_t zm) __a
 
 // CHECK-LABEL: @test_smlal_lane_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlal_lane_x2_s8j10svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -1374,16 +1010,12 @@ void test_smlal_lane_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __ar
 
 // CHECK-LABEL: @test_smlal_lane_x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlal_lane_x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -1393,16 +1025,12 @@ void test_smlal_lane_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) _
 
 // CHECK-LABEL: @test_smlal_lane_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlal_lane_x2_u8j11svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -1412,16 +1040,12 @@ void test_smlal_lane_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __
 
 // CHECK-LABEL: @test_smlal_lane_x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlal_lane_x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -1433,16 +1057,12 @@ void test_smlal_lane_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_smlsl_lane_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlsl_lane_x2_s8j10svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -1452,16 +1072,12 @@ void test_smlsl_lane_x2_s8(uint32_t slice_base, svint8x2_t zn, svint8_t zm) __ar
 
 // CHECK-LABEL: @test_smlsl_lane_x2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlsl_lane_x2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -1471,16 +1087,12 @@ void test_smlsl_lane_x2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) _
 
 // CHECK-LABEL: @test_smlsl_lane_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlsl_lane_x2_u8j11svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -1490,16 +1102,12 @@ void test_smlsl_lane_x2_u8(uint32_t slice_base, svuint8x2_t zn, svuint8_t zm) __
 
 // CHECK-LABEL: @test_smlsl_lane_x2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlsl_lane_x2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -1511,16 +1119,12 @@ void test_smlsl_lane_x2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_sumlall_lane_x2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_sumlall_lane_x2_s8j10svint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_sumlall_lane_x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -1530,16 +1134,12 @@ void test_sumlall_lane_x2_s8(uint32_t slice_base, svint8x2_t zn, svuint8_t zm) _
 
 // CHECK-LABEL: @test_usmlall_lane_x2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_usmlall_lane_x2_u8j11svuint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x2.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_usmlall_lane_x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -1555,20 +1155,12 @@ void test_usmlall_lane_x2_u8(uint32_t slice_base, svuint8x2_t zn, svint8_t zm) _
 
 // CHECK-LABEL: @test_smlal_lane_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlal_lane_x4_s8j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -1578,20 +1170,12 @@ void test_smlal_lane_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __ar
 
 // CHECK-LABEL: @test_smlal_lane_x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlal_lane_x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -1601,20 +1185,12 @@ void test_smlal_lane_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) _
 
 // CHECK-LABEL: @test_smlal_lane_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlal_lane_x4_u8j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -1624,20 +1200,12 @@ void test_smlal_lane_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __
 
 // CHECK-LABEL: @test_smlal_lane_x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlal_lane_x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umla.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlal_lane_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -1649,20 +1217,12 @@ void test_smlal_lane_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_smlsl_lane_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlsl_lane_x4_s8j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za")
@@ -1672,20 +1232,12 @@ void test_smlsl_lane_x4_s8(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __ar
 
 // CHECK-LABEL: @test_smlsl_lane_x4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlsl_lane_x4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -1695,20 +1247,12 @@ void test_smlsl_lane_x4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) _
 
 // CHECK-LABEL: @test_smlsl_lane_x4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_smlsl_lane_x4_u8j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -1718,20 +1262,12 @@ void test_smlsl_lane_x4_u8(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __
 
 // CHECK-LABEL: @test_smlsl_lane_x4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_smlsl_lane_x4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umls.za64.lane.vg4x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_smlsl_lane_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -1743,20 +1279,12 @@ void test_smlsl_lane_x4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_sumlall_lane_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_sumlall_lane_x4_s8j10svint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sumla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_sumlall_lane_x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za")
@@ -1768,20 +1296,12 @@ void test_sumlall_lane_x4_s8(uint32_t slice_base, svint8x4_t zn, svuint8_t zm) _
 
 // CHECK-LABEL: @test_usmlall_lane_x4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_usmlall_lane_x4_s8j11svuint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usmla.za32.lane.vg4x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 15)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_usmlall_lane_x4_s8(uint32_t slice_base, svuint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za")

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mls.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mls.c
index 1986317ca719dd..2b1554cd9d8b08 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mls.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mls.c
@@ -19,20 +19,12 @@
 // Multi, multi
 // CHECK-LABEL: @test_svmls2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls2_f32j13svfloat32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32x2_t zm) __arm_streaming __arm_inout("za") {
@@ -41,28 +33,12 @@ void test_svmls2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32x2_t zm) __
 
 // CHECK-LABEL: @test_svmls4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls4_f32j13svfloat32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE2:%.*]], <vscale x 4 x float> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32x4_t zm) __arm_streaming __arm_inout("za") {
@@ -73,16 +49,12 @@ void test_svmls4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32x4_t zm) __
 // Multi, single
 // CHECK-LABEL: @test_svmls_single2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single2_f32j13svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -91,20 +63,12 @@ void test_svmls_single2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t z
 
 // CHECK-LABEL: @test_svmls_single4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single4_f32j13svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -115,16 +79,12 @@ void test_svmls_single4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t z
 // Multi, indexed
 // CHECK-LABEL: @test_svmls_lane2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane2_f32j13svfloat32x2_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -133,20 +93,12 @@ void test_svmls_lane2_f32(uint32_t slice_base, svfloat32x2_t zn, svfloat32_t zm)
 
 // CHECK-LABEL: @test_svmls_lane4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane4_f32j13svfloat32x4_tu13__SVFloat32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]], <vscale x 4 x float> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t zm) __arm_streaming __arm_inout("za") {
@@ -157,20 +109,12 @@ void test_svmls_lane4_f32(uint32_t slice_base, svfloat32x4_t zn, svfloat32_t zm)
 // Multi, multi
 // CHECK-LABEL: @test_svmls2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls2_f64j13svfloat64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64x2_t zm) __arm_streaming __arm_inout("za") {
@@ -179,28 +123,12 @@ void test_svmls2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64x2_t zm) __
 
 // CHECK-LABEL: @test_svmls4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls4_f64j13svfloat64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE2:%.*]], <vscale x 2 x double> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64x4_t zm) __arm_streaming __arm_inout("za") {
@@ -211,16 +139,12 @@ void test_svmls4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64x4_t zm) __
 // Multi, single
 // CHECK-LABEL: @test_svmls_single2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single2_f64j13svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {
@@ -229,20 +153,12 @@ void test_svmls_single2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t z
 
 // CHECK-LABEL: @test_svmls_single4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single4_f64j13svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.single.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {
@@ -253,16 +169,12 @@ void test_svmls_single4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64_t z
 // Multi, indexed
 // CHECK-LABEL: @test_svmls_lane2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane2_f64j13svfloat64x2_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {
@@ -271,20 +183,12 @@ void test_svmls_lane2_f64(uint32_t slice_base, svfloat64x2_t zn, svfloat64_t zm)
 
 // CHECK-LABEL: @test_svmls_lane4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane4_f64j13svfloat64x4_tu13__SVFloat64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmls.lane.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]], <vscale x 2 x double> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane4_f64(uint32_t slice_base, svfloat64x4_t zn, svfloat64_t zm) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlsl.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlsl.c
index 121d0fad2ae9bd..e56ffaa1db03e5 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlsl.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_mlsl.c
@@ -19,20 +19,12 @@
 // Multi, multi
 // CHECK-LABEL: @test_svmls2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls2_f16j13svfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16x2_t zm) __arm_streaming __arm_inout("za")
@@ -42,20 +34,12 @@ void test_svmls2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16x2_t zm) __
 
 // CHECK-LABEL: @test_svmls2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z16test_svmls2_bf16j14svbfloat16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16x2_t zm) __arm_streaming __arm_inout("za")
@@ -65,20 +49,12 @@ void test_svmls2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16x2_t zm)
 
 // CHECK-LABEL: @test_svmls2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls2_u16j12svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -88,20 +64,12 @@ void test_svmls2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16x2_t zm) __ar
 
 // CHECK-LABEL: @test_svmls2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls2_s16j11svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_streaming __arm_inout("za")
@@ -111,28 +79,12 @@ void test_svmls2_s16(uint32_t slice_base, svint16x2_t zn, svint16x2_t zm) __arm_
 
 // CHECK-LABEL: @test_svmls4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls4_f16j13svfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE2:%.*]], <vscale x 8 x half> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16x4_t zm) __arm_streaming __arm_inout("za")
@@ -142,28 +94,12 @@ void test_svmls4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16x4_t zm) __
 
 // CHECK-LABEL: @test_svmls4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z16test_svmls4_bf16j14svbfloat16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16x4_t zm) __arm_streaming __arm_inout("za")
@@ -173,28 +109,12 @@ void test_svmls4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16x4_t zm)
 
 // CHECK-LABEL: @test_svmls4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls4_u16j12svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -204,28 +124,12 @@ void test_svmls4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16x4_t zm) __ar
 
 // CHECK-LABEL: @test_svmls4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z15test_svmls4_s16j11svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls4_s16(uint32_t slice_base, svint16x4_t zn, svint16x4_t zm) __arm_streaming __arm_inout("za")
@@ -297,16 +201,12 @@ void test_svmls_single1_s16(uint32_t slice_base, svint16_t zn, svint16_t zm) __a
 
 // CHECK-LABEL: @test_svmls_single2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single2_f16j13svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -316,16 +216,12 @@ void test_svmls_single2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t z
 
 // CHECK-LABEL: @test_svmls_single2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmls_single2_bf16j14svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -335,16 +231,12 @@ void test_svmls_single2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_
 
 // CHECK-LABEL: @test_svmls_single2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -354,16 +246,12 @@ void test_svmls_single2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_svmls_single2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -373,20 +261,12 @@ void test_svmls_single2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) _
 
 // CHECK-LABEL: @test_svmls_single4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single4_f16j13svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -396,20 +276,12 @@ void test_svmls_single4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t z
 
 // CHECK-LABEL: @test_svmls_single4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z23test_svmls_single4_bf16j14svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.single.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -419,20 +291,12 @@ void test_svmls_single4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_
 
 // CHECK-LABEL: @test_svmls_single4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -442,20 +306,12 @@ void test_svmls_single4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm)
 
 // CHECK-LABEL: @test_svmls_single4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z22test_svmls_single4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.single.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_single4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -529,16 +385,12 @@ void test_svmls_lane1_s16(uint32_t slice_base, svint16_t zn, svint16_t zm) __arm
 
 // CHECK-LABEL: @test_svmls_lane2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane2_f16j13svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -548,16 +400,12 @@ void test_svmls_lane2_f16(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm)
 
 // CHECK-LABEL: @test_svmls_lane2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_svmls_lane2_bf16j14svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -567,16 +415,12 @@ void test_svmls_lane2_bf16(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t
 
 // CHECK-LABEL: @test_svmls_lane2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane2_u16j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -586,16 +430,12 @@ void test_svmls_lane2_u16(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) _
 
 // CHECK-LABEL: @test_svmls_lane2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane2_s16j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za")
@@ -605,20 +445,12 @@ void test_svmls_lane2_s16(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __a
 
 // CHECK-LABEL: @test_svmls_lane4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane4_f16j13svfloat16x4_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -628,20 +460,12 @@ void test_svmls_lane4_f16(uint32_t slice_base, svfloat16x4_t zn, svfloat16_t zm)
 
 // CHECK-LABEL: @test_svmls_lane4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z21test_svmls_lane4_bf16j14svbfloat16x4_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fmlsl.lane.vg2x4.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za")
@@ -651,20 +475,12 @@ void test_svmls_lane4_bf16(uint32_t slice_base, svbfloat16x4_t zn, svbfloat16_t
 
 // CHECK-LABEL: @test_svmls_lane4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane4_u16j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.umlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za")
@@ -674,20 +490,12 @@ void test_svmls_lane4_u16(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) _
 
 // CHECK-LABEL: @test_svmls_lane4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z20test_svmls_lane4_s16j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.smlsl.lane.vg2x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svmls_lane4_s16(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za")

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_read.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_read.c
index e042161262b33b..da17c6b13d17c8 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_read.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_read.c
@@ -9,21 +9,27 @@
 
 // CHECK-LABEL: @test_svread_ver_za8_u8_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_ver_za8_u8_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svread_ver_za8_u8_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za8_u8_vg2(0, base);
@@ -31,21 +37,27 @@ svuint8x2_t test_svread_ver_za8_u8_vg2(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_ver_za8_s8_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_ver_za8_s8_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svread_ver_za8_s8_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za8_s8_vg2(0, base);
@@ -53,21 +65,27 @@ svint8x2_t test_svread_ver_za8_s8_vg2(uint32_t base) __arm_streaming __arm_in("z
 
 // CHECK-LABEL: @test_svread_hor_za8_u8_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_hor_za8_u8_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svread_hor_za8_u8_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za8_u8_vg2(0, base);
@@ -75,21 +93,27 @@ svuint8x2_t test_svread_hor_za8_u8_vg2(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_hor_za8_s8_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_hor_za8_s8_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svread_hor_za8_s8_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za8_s8_vg2(0, base);
@@ -97,6 +121,7 @@ svint8x2_t test_svread_hor_za8_s8_vg2(uint32_t base) __arm_streaming __arm_in("z
 
 // CHECK-LABEL: @test_svread_hor_za8_u8_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -106,10 +131,13 @@ svint8x2_t test_svread_hor_za8_s8_vg2(uint32_t base) __arm_streaming __arm_in("z
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_hor_za8_u8_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -119,7 +147,9 @@ svint8x2_t test_svread_hor_za8_s8_vg2(uint32_t base) __arm_streaming __arm_in("z
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svread_hor_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za8_u8_vg4(0, base);
@@ -127,6 +157,7 @@ svuint8x4_t test_svread_hor_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_hor_za8_s8_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -136,10 +167,13 @@ svuint8x4_t test_svread_hor_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_hor_za8_s8_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -149,7 +183,9 @@ svuint8x4_t test_svread_hor_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svread_hor_za8_s8_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za8_s8_vg4(0, base);
@@ -157,6 +193,7 @@ svint8x4_t test_svread_hor_za8_s8_vg4(uint32_t base) __arm_streaming __arm_in("z
 
 // CHECK-LABEL: @test_svread_ver_za8_u8_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -166,10 +203,13 @@ svint8x4_t test_svread_hor_za8_s8_vg4(uint32_t base) __arm_streaming __arm_in("z
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_ver_za8_u8_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -179,7 +219,9 @@ svint8x4_t test_svread_hor_za8_s8_vg4(uint32_t base) __arm_streaming __arm_in("z
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svread_ver_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za8_u8_vg4(0, base);
@@ -187,6 +229,7 @@ svuint8x4_t test_svread_ver_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_ver_za8_s8_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -196,10 +239,13 @@ svuint8x4_t test_svread_ver_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_ver_za8_s8_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -209,7 +255,9 @@ svuint8x4_t test_svread_ver_za8_u8_vg4(uint32_t base) __arm_streaming __arm_in("
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svread_ver_za8_s8_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za8_s8_vg4(0, base);
@@ -217,21 +265,27 @@ svint8x4_t test_svread_ver_za8_s8_vg4(uint32_t base) __arm_streaming __arm_in("z
 
 // CHECK-LABEL: @test_svread_hor_za16_u16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za16_u16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svread_hor_za16_u16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_u16_vg2(1, base);
@@ -239,21 +293,27 @@ svuint16x2_t test_svread_hor_za16_u16_vg2(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_hor_za16_bf16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.hor.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svread_hor_za16_bf16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.hor.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svread_hor_za16_bf16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_bf16_vg2(1, base);
@@ -261,21 +321,27 @@ svbfloat16x2_t test_svread_hor_za16_bf16_vg2(uint32_t base) __arm_streaming __ar
 
 // CHECK-LABEL: @test_svread_hor_za16_f16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.hor.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za16_f16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.hor.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svread_hor_za16_f16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_f16_vg2(1, base);
@@ -283,21 +349,27 @@ svfloat16x2_t test_svread_hor_za16_f16_vg2(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_hor_za16_s16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za16_s16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svread_hor_za16_s16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_s16_vg2(1, base);
@@ -305,21 +377,27 @@ svint16x2_t test_svread_hor_za16_s16_vg2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_ver_za16_u16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za16_u16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svread_ver_za16_u16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_u16_vg2(1, base);
@@ -327,21 +405,27 @@ svuint16x2_t test_svread_ver_za16_u16_vg2(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_ver_za16_bf16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.ver.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svread_ver_za16_bf16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.ver.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svread_ver_za16_bf16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_bf16_vg2(1, base);
@@ -349,21 +433,27 @@ svbfloat16x2_t test_svread_ver_za16_bf16_vg2(uint32_t base) __arm_streaming __ar
 
 // CHECK-LABEL: @test_svread_ver_za16_f16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.ver.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za16_f16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.ver.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svread_ver_za16_f16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_f16_vg2(1, base);
@@ -371,21 +461,27 @@ svfloat16x2_t test_svread_ver_za16_f16_vg2(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_ver_za16_s16_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za16_s16_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svread_ver_za16_s16_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_s16_vg2(1, base);
@@ -393,6 +489,7 @@ svint16x2_t test_svread_ver_za16_s16_vg2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_hor_za16_u16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -402,10 +499,13 @@ svint16x2_t test_svread_ver_za16_s16_vg2(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za16_u16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -415,7 +515,9 @@ svint16x2_t test_svread_ver_za16_s16_vg2(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svread_hor_za16_u16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_u16_vg4(1, base);
@@ -423,6 +525,7 @@ svuint16x4_t test_svread_hor_za16_u16_vg4(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_hor_za16_bf16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.hor.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -432,10 +535,13 @@ svuint16x4_t test_svread_hor_za16_u16_vg4(uint32_t base) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svread_hor_za16_bf16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.hor.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -445,7 +551,9 @@ svuint16x4_t test_svread_hor_za16_u16_vg4(uint32_t base) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svread_hor_za16_bf16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_bf16_vg4(1, base);
@@ -453,6 +561,7 @@ svbfloat16x4_t test_svread_hor_za16_bf16_vg4(uint32_t base) __arm_streaming __ar
 
 // CHECK-LABEL: @test_svread_hor_za16_f16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.hor.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -462,10 +571,13 @@ svbfloat16x4_t test_svread_hor_za16_bf16_vg4(uint32_t base) __arm_streaming __ar
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za16_f16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.hor.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -475,7 +587,9 @@ svbfloat16x4_t test_svread_hor_za16_bf16_vg4(uint32_t base) __arm_streaming __ar
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svread_hor_za16_f16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_f16_vg4(1, base);
@@ -483,6 +597,7 @@ svfloat16x4_t test_svread_hor_za16_f16_vg4(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_hor_za16_s16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -492,10 +607,13 @@ svfloat16x4_t test_svread_hor_za16_f16_vg4(uint32_t base) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za16_s16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -505,7 +623,9 @@ svfloat16x4_t test_svread_hor_za16_f16_vg4(uint32_t base) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svread_hor_za16_s16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za16_s16_vg4(1, base);
@@ -513,6 +633,7 @@ svint16x4_t test_svread_hor_za16_s16_vg4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_ver_za16_u16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -522,10 +643,13 @@ svint16x4_t test_svread_hor_za16_s16_vg4(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za16_u16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -535,7 +659,9 @@ svint16x4_t test_svread_hor_za16_s16_vg4(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svread_ver_za16_u16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_u16_vg4(1, base);
@@ -543,6 +669,7 @@ svuint16x4_t test_svread_ver_za16_u16_vg4(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_ver_za16_bf16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.ver.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -552,10 +679,13 @@ svuint16x4_t test_svread_ver_za16_u16_vg4(uint32_t base) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svread_ver_za16_bf16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.ver.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -565,7 +695,9 @@ svuint16x4_t test_svread_ver_za16_u16_vg4(uint32_t base) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svread_ver_za16_bf16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_bf16_vg4(1, base);
@@ -573,6 +705,7 @@ svbfloat16x4_t test_svread_ver_za16_bf16_vg4(uint32_t base) __arm_streaming __ar
 
 // CHECK-LABEL: @test_svread_ver_za16_f16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.ver.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -582,10 +715,13 @@ svbfloat16x4_t test_svread_ver_za16_bf16_vg4(uint32_t base) __arm_streaming __ar
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za16_f16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.ver.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -595,7 +731,9 @@ svbfloat16x4_t test_svread_ver_za16_bf16_vg4(uint32_t base) __arm_streaming __ar
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svread_ver_za16_f16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_f16_vg4(1, base);
@@ -603,6 +741,7 @@ svfloat16x4_t test_svread_ver_za16_f16_vg4(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_ver_za16_s16_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -612,10 +751,13 @@ svfloat16x4_t test_svread_ver_za16_f16_vg4(uint32_t base) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za16_s16_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -625,7 +767,9 @@ svfloat16x4_t test_svread_ver_za16_f16_vg4(uint32_t base) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svread_ver_za16_s16_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za16_s16_vg4(1, base);
@@ -633,21 +777,27 @@ svint16x4_t test_svread_ver_za16_s16_vg4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_hor_za32_u32_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za32_u32_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svread_hor_za32_u32_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za32_u32_vg2(3, base);
@@ -655,21 +805,27 @@ svuint32x2_t test_svread_hor_za32_u32_vg2(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_hor_za32_f32_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.hor.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za32_f32_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.hor.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svread_hor_za32_f32_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za32_f32_vg2(3, base);
@@ -677,21 +833,27 @@ svfloat32x2_t test_svread_hor_za32_f32_vg2(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_hor_za32_s32_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za32_s32_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svread_hor_za32_s32_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za32_s32_vg2(3, base);
@@ -699,21 +861,27 @@ svint32x2_t test_svread_hor_za32_s32_vg2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_ver_za32_u32_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za32_u32_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svread_ver_za32_u32_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za32_u32_vg2(3, base);
@@ -721,21 +889,27 @@ svuint32x2_t test_svread_ver_za32_u32_vg2(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_ver_za32_f32_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.ver.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za32_f32_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.ver.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svread_ver_za32_f32_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za32_f32_vg2(3, base);
@@ -743,21 +917,27 @@ svfloat32x2_t test_svread_ver_za32_f32_vg2(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_ver_za32_s32_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za32_s32_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svread_ver_za32_s32_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za32_s32_vg2(3, base);
@@ -765,6 +945,7 @@ svint32x2_t test_svread_ver_za32_s32_vg2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_hor_za32_u32_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -774,10 +955,13 @@ svint32x2_t test_svread_ver_za32_s32_vg2(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za32_u32_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -787,7 +971,9 @@ svint32x2_t test_svread_ver_za32_s32_vg2(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svread_hor_za32_u32_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za32_u32_vg4(3, base);
@@ -795,6 +981,7 @@ svuint32x4_t test_svread_hor_za32_u32_vg4(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_hor_za32_f32_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.hor.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -804,10 +991,13 @@ svuint32x4_t test_svread_hor_za32_u32_vg4(uint32_t base) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za32_f32_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.hor.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -817,7 +1007,9 @@ svuint32x4_t test_svread_hor_za32_u32_vg4(uint32_t base) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svread_hor_za32_f32_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za32_f32_vg4(3, base);
@@ -825,6 +1017,7 @@ svfloat32x4_t test_svread_hor_za32_f32_vg4(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_hor_za32_s32_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -834,10 +1027,13 @@ svfloat32x4_t test_svread_hor_za32_f32_vg4(uint32_t base) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za32_s32_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -847,7 +1043,9 @@ svfloat32x4_t test_svread_hor_za32_f32_vg4(uint32_t base) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svread_hor_za32_s32_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za32_s32_vg4(3, base);
@@ -855,6 +1053,7 @@ svint32x4_t test_svread_hor_za32_s32_vg4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_ver_za32_u32_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -864,10 +1063,13 @@ svint32x4_t test_svread_hor_za32_s32_vg4(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za32_u32_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -877,7 +1079,9 @@ svint32x4_t test_svread_hor_za32_s32_vg4(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svread_ver_za32_u32_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za32_u32_vg4(3, base);
@@ -885,6 +1089,7 @@ svuint32x4_t test_svread_ver_za32_u32_vg4(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_ver_za32_f32_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.ver.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -894,10 +1099,13 @@ svuint32x4_t test_svread_ver_za32_u32_vg4(uint32_t base) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za32_f32_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.ver.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -907,7 +1115,9 @@ svuint32x4_t test_svread_ver_za32_u32_vg4(uint32_t base) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svread_ver_za32_f32_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za32_f32_vg4(3, base);
@@ -915,6 +1125,7 @@ svfloat32x4_t test_svread_ver_za32_f32_vg4(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_ver_za32_s32_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -924,10 +1135,13 @@ svfloat32x4_t test_svread_ver_za32_f32_vg4(uint32_t base) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za32_s32_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -937,7 +1151,9 @@ svfloat32x4_t test_svread_ver_za32_f32_vg4(uint32_t base) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svread_ver_za32_s32_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za32_s32_vg4(3, base);
@@ -945,21 +1161,27 @@ svint32x4_t test_svread_ver_za32_s32_vg4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_hor_za64_u64_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za64_u64_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svread_hor_za64_u64_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za64_u64_vg2(7, base);
@@ -967,21 +1189,27 @@ svuint64x2_t test_svread_hor_za64_u64_vg2(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_hor_za64_f64_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.hor.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za64_f64_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.hor.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svread_hor_za64_f64_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za64_f64_vg2(7, base);
@@ -989,21 +1217,27 @@ svfloat64x2_t test_svread_hor_za64_f64_vg2(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_hor_za64_s64_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za64_s64_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svread_hor_za64_s64_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za64_s64_vg2(7, base);
@@ -1011,21 +1245,27 @@ svint64x2_t test_svread_hor_za64_s64_vg2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_ver_za64_u64_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za64_u64_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svread_ver_za64_u64_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za64_u64_vg2(7, base);
@@ -1033,21 +1273,27 @@ svuint64x2_t test_svread_ver_za64_u64_vg2(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_ver_za64_f64_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.ver.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za64_f64_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.ver.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svread_ver_za64_f64_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za64_f64_vg2(7, base);
@@ -1055,21 +1301,27 @@ svfloat64x2_t test_svread_ver_za64_f64_vg2(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_ver_za64_s64_vg2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za64_s64_vg2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svread_ver_za64_s64_vg2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za64_s64_vg2(7, base);
@@ -1077,6 +1329,7 @@ svint64x2_t test_svread_ver_za64_s64_vg2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_hor_za64_u64_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1086,10 +1339,13 @@ svint64x2_t test_svread_ver_za64_s64_vg2(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za64_u64_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1099,7 +1355,9 @@ svint64x2_t test_svread_ver_za64_s64_vg2(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svread_hor_za64_u64_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za64_u64_vg4(7, base);
@@ -1107,6 +1365,7 @@ svuint64x4_t test_svread_hor_za64_u64_vg4(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_hor_za64_f64_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.hor.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1116,10 +1375,13 @@ svuint64x4_t test_svread_hor_za64_u64_vg4(uint32_t base) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za64_f64_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.hor.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1129,7 +1391,9 @@ svuint64x4_t test_svread_hor_za64_u64_vg4(uint32_t base) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svread_hor_za64_f64_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za64_f64_vg4(7, base);
@@ -1137,6 +1401,7 @@ svfloat64x4_t test_svread_hor_za64_f64_vg4(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_hor_za64_s64_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1146,10 +1411,13 @@ svfloat64x4_t test_svread_hor_za64_f64_vg4(uint32_t base) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_hor_za64_s64_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1159,7 +1427,9 @@ svfloat64x4_t test_svread_hor_za64_f64_vg4(uint32_t base) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svread_hor_za64_s64_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_hor_za64_s64_vg4(7, base);
@@ -1167,6 +1437,7 @@ svint64x4_t test_svread_hor_za64_s64_vg4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_ver_za64_u64_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1176,10 +1447,13 @@ svint64x4_t test_svread_hor_za64_s64_vg4(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za64_u64_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1189,7 +1463,9 @@ svint64x4_t test_svread_hor_za64_s64_vg4(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svread_ver_za64_u64_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za64_u64_vg4(7, base);
@@ -1197,6 +1473,7 @@ svuint64x4_t test_svread_ver_za64_u64_vg4(uint32_t base) __arm_streaming __arm_i
 
 // CHECK-LABEL: @test_svread_ver_za64_f64_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.ver.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1206,10 +1483,13 @@ svuint64x4_t test_svread_ver_za64_u64_vg4(uint32_t base) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za64_f64_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.ver.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1219,7 +1499,9 @@ svuint64x4_t test_svread_ver_za64_u64_vg4(uint32_t base) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svread_ver_za64_f64_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za64_f64_vg4(7, base);
@@ -1227,6 +1509,7 @@ svfloat64x4_t test_svread_ver_za64_f64_vg4(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_ver_za64_s64_vg4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1236,10 +1519,13 @@ svfloat64x4_t test_svread_ver_za64_f64_vg4(uint32_t base) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svread_ver_za64_s64_vg4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1249,7 +1535,9 @@ svfloat64x4_t test_svread_ver_za64_f64_vg4(uint32_t base) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svread_ver_za64_s64_vg4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_ver_za64_s64_vg4(7, base);
@@ -1257,21 +1545,27 @@ svint64x4_t test_svread_ver_za64_s64_vg4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_za8_s8_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x2.nxv16i8(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svread_za8_s8_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x2.nxv16i8(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svread_za8_s8_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za8_s8_vg1x2(base);
@@ -1279,21 +1573,27 @@ svint8x2_t test_svread_za8_s8_vg1x2(uint32_t base) __arm_streaming __arm_in("za"
 
 // CHECK-LABEL: @test_svread_za8_u8_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x2.nxv16i8(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svread_za8_u8_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x2.nxv16i8(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svread_za8_u8_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za8_u8_vg1x2(base);
@@ -1301,43 +1601,56 @@ svuint8x2_t test_svread_za8_u8_vg1x2(uint32_t base) __arm_streaming __arm_in("za
 
 // CHECK-LABEL: @test_svread_za16_s16_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x2.nxv8i16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za16_s16_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x2.nxv8i16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svread_za16_s16_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_s16_vg1x2(base);
 }
 
+//
 // CHECK-LABEL: @test_svread_za16_u16_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x2.nxv8i16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za16_u16_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x2.nxv8i16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svread_za16_u16_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_u16_vg1x2(base);
@@ -1345,21 +1658,27 @@ svuint16x2_t test_svread_za16_u16_vg1x2(uint32_t base) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svread_za16_bf16_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.vg1x2.nxv8bf16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svread_za16_bf16_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.vg1x2.nxv8bf16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svread_za16_bf16_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_bf16_vg1x2(base);
@@ -1367,21 +1686,27 @@ svbfloat16x2_t test_svread_za16_bf16_vg1x2(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_za16_f16_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.vg1x2.nxv8f16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za16_f16_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.vg1x2.nxv8f16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svread_za16_f16_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_f16_vg1x2(base);
@@ -1389,21 +1714,27 @@ svfloat16x2_t test_svread_za16_f16_vg1x2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_za32_s32_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x2.nxv4i32(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za32_s32_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x2.nxv4i32(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svread_za32_s32_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za32_s32_vg1x2(base);
@@ -1411,21 +1742,27 @@ svint32x2_t test_svread_za32_s32_vg1x2(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_za32_u32_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x2.nxv4i32(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za32_u32_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x2.nxv4i32(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svread_za32_u32_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za32_u32_vg1x2(base);
@@ -1433,21 +1770,27 @@ svuint32x2_t test_svread_za32_u32_vg1x2(uint32_t base) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svread_za32_f32_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.vg1x2.nxv4f32(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za32_f32_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.vg1x2.nxv4f32(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svread_za32_f32_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za32_f32_vg1x2(base);
@@ -1455,21 +1798,27 @@ svfloat32x2_t test_svread_za32_f32_vg1x2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_za64_u64_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x2.nxv2i64(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za64_u64_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x2.nxv2i64(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svread_za64_u64_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za64_u64_vg1x2(base);
@@ -1477,21 +1826,27 @@ svuint64x2_t test_svread_za64_u64_vg1x2(uint32_t base) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svread_za64_f64_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.vg1x2.nxv2f64(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za64_f64_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.vg1x2.nxv2f64(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svread_za64_f64_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za64_f64_vg1x2(base);
@@ -1499,21 +1854,27 @@ svfloat64x2_t test_svread_za64_f64_vg1x2(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_za64_s64_vg1x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x2.nxv2i64(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za64_s64_vg1x2j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x2.nxv2i64(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svread_za64_s64_vg1x2(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za64_s64_vg1x2(base);
@@ -1521,6 +1882,7 @@ svint64x2_t test_svread_za64_s64_vg1x2(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_za8_s8_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x4.nxv16i8(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1530,10 +1892,13 @@ svint64x2_t test_svread_za64_s64_vg1x2(uint32_t base) __arm_streaming __arm_in("
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svread_za8_s8_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x4.nxv16i8(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1543,7 +1908,9 @@ svint64x2_t test_svread_za64_s64_vg1x2(uint32_t base) __arm_streaming __arm_in("
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svread_za8_s8_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za8_s8_vg1x4(base);
@@ -1551,6 +1918,7 @@ svint8x4_t test_svread_za8_s8_vg1x4(uint32_t base) __arm_streaming __arm_in("za"
 
 // CHECK-LABEL: @test_svread_za8_u8_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x4.nxv16i8(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1560,10 +1928,13 @@ svint8x4_t test_svread_za8_s8_vg1x4(uint32_t base) __arm_streaming __arm_in("za"
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svread_za8_u8_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.read.vg1x4.nxv16i8(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1573,7 +1944,9 @@ svint8x4_t test_svread_za8_s8_vg1x4(uint32_t base) __arm_streaming __arm_in("za"
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svread_za8_u8_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za8_u8_vg1x4(base);
@@ -1581,6 +1954,7 @@ svuint8x4_t test_svread_za8_u8_vg1x4(uint32_t base) __arm_streaming __arm_in("za
 
 // CHECK-LABEL: @test_svread_za16_s16_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x4.nxv8i16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1590,10 +1964,13 @@ svuint8x4_t test_svread_za8_u8_vg1x4(uint32_t base) __arm_streaming __arm_in("za
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za16_s16_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x4.nxv8i16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1603,7 +1980,9 @@ svuint8x4_t test_svread_za8_u8_vg1x4(uint32_t base) __arm_streaming __arm_in("za
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svread_za16_s16_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_s16_vg1x4(base);
@@ -1611,6 +1990,7 @@ svint16x4_t test_svread_za16_s16_vg1x4(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_za16_u16_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x4.nxv8i16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1620,10 +2000,13 @@ svint16x4_t test_svread_za16_s16_vg1x4(uint32_t base) __arm_streaming __arm_in("
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za16_u16_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.read.vg1x4.nxv8i16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1633,7 +2016,9 @@ svint16x4_t test_svread_za16_s16_vg1x4(uint32_t base) __arm_streaming __arm_in("
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svread_za16_u16_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_u16_vg1x4(base);
@@ -1641,6 +2026,7 @@ svuint16x4_t test_svread_za16_u16_vg1x4(uint32_t base) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svread_za16_bf16_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.vg1x4.nxv8bf16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -1650,10 +2036,13 @@ svuint16x4_t test_svread_za16_u16_vg1x4(uint32_t base) __arm_streaming __arm_in(
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svread_za16_bf16_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.read.vg1x4.nxv8bf16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -1663,7 +2052,9 @@ svuint16x4_t test_svread_za16_u16_vg1x4(uint32_t base) __arm_streaming __arm_in(
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svread_za16_bf16_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_bf16_vg1x4(base);
@@ -1671,6 +2062,7 @@ svbfloat16x4_t test_svread_za16_bf16_vg1x4(uint32_t base) __arm_streaming __arm_
 
 // CHECK-LABEL: @test_svread_za16_f16_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.vg1x4.nxv8f16(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -1680,10 +2072,13 @@ svbfloat16x4_t test_svread_za16_bf16_vg1x4(uint32_t base) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za16_f16_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.read.vg1x4.nxv8f16(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -1693,7 +2088,9 @@ svbfloat16x4_t test_svread_za16_bf16_vg1x4(uint32_t base) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svread_za16_f16_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za16_f16_vg1x4(base);
@@ -1701,6 +2098,7 @@ svfloat16x4_t test_svread_za16_f16_vg1x4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_za32_s32_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x4.nxv4i32(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1710,10 +2108,13 @@ svfloat16x4_t test_svread_za16_f16_vg1x4(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za32_s32_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x4.nxv4i32(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1723,7 +2124,9 @@ svfloat16x4_t test_svread_za16_f16_vg1x4(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svread_za32_s32_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za32_s32_vg1x4(base);
@@ -1731,6 +2134,7 @@ svint32x4_t test_svread_za32_s32_vg1x4(uint32_t base) __arm_streaming __arm_in("
 
 // CHECK-LABEL: @test_svread_za32_u32_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x4.nxv4i32(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1740,10 +2144,13 @@ svint32x4_t test_svread_za32_s32_vg1x4(uint32_t base) __arm_streaming __arm_in("
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za32_u32_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.read.vg1x4.nxv4i32(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1753,7 +2160,9 @@ svint32x4_t test_svread_za32_s32_vg1x4(uint32_t base) __arm_streaming __arm_in("
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svread_za32_u32_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za32_u32_vg1x4(base);
@@ -1761,6 +2170,7 @@ svuint32x4_t test_svread_za32_u32_vg1x4(uint32_t base) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svread_za32_f32_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.vg1x4.nxv4f32(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -1770,10 +2180,13 @@ svuint32x4_t test_svread_za32_u32_vg1x4(uint32_t base) __arm_streaming __arm_in(
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za32_f32_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.read.vg1x4.nxv4f32(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -1783,7 +2196,9 @@ svuint32x4_t test_svread_za32_u32_vg1x4(uint32_t base) __arm_streaming __arm_in(
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svread_za32_f32_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za32_f32_vg1x4(base);
@@ -1791,6 +2206,7 @@ svfloat32x4_t test_svread_za32_f32_vg1x4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_za64_u64_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x4.nxv2i64(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1800,10 +2216,13 @@ svfloat32x4_t test_svread_za32_f32_vg1x4(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za64_u64_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x4.nxv2i64(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1813,7 +2232,9 @@ svfloat32x4_t test_svread_za32_f32_vg1x4(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svread_za64_u64_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za64_u64_vg1x4(base);
@@ -1821,6 +2242,7 @@ svuint64x4_t test_svread_za64_u64_vg1x4(uint32_t base) __arm_streaming __arm_in(
 
 // CHECK-LABEL: @test_svread_za64_f64_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.vg1x4.nxv2f64(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1830,10 +2252,13 @@ svuint64x4_t test_svread_za64_u64_vg1x4(uint32_t base) __arm_streaming __arm_in(
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za64_f64_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.read.vg1x4.nxv2f64(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1843,7 +2268,9 @@ svuint64x4_t test_svread_za64_u64_vg1x4(uint32_t base) __arm_streaming __arm_in(
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svread_za64_f64_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za64_f64_vg1x4(base);
@@ -1851,6 +2278,7 @@ svfloat64x4_t test_svread_za64_f64_vg1x4(uint32_t base) __arm_streaming __arm_in
 
 // CHECK-LABEL: @test_svread_za64_s64_vg1x4(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x4.nxv2i64(i32 [[BASE:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1860,10 +2288,13 @@ svfloat64x4_t test_svread_za64_f64_vg1x4(uint32_t base) __arm_streaming __arm_in
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svread_za64_s64_vg1x4j(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.read.vg1x4.nxv2i64(i32 [[BASE:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1873,7 +2304,9 @@ svfloat64x4_t test_svread_za64_f64_vg1x4(uint32_t base) __arm_streaming __arm_in
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svread_za64_s64_vg1x4(uint32_t base) __arm_streaming __arm_in("za") {
   return svread_za64_s64_vg1x4(base);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sqdmulh.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sqdmulh.c
index 195531cf131e38..26804866a7563c 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sqdmulh.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sqdmulh.c
@@ -19,25 +19,27 @@
 
 // CHECK-LABEL: @test_svqdmulh_single_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svqdmulh_single_s8_x210svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svqdmulh_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s8_x2,,,)(zdn, zm);
@@ -45,25 +47,27 @@ svint8x2_t test_svqdmulh_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streami
 
 // CHECK-LABEL: @test_svqdmulh_single_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svqdmulh_single_s16_x211svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svqdmulh_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s16_x2,,,)(zdn, zm);
@@ -71,25 +75,27 @@ svint16x2_t test_svqdmulh_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_str
 
 // CHECK-LABEL: @test_svqdmulh_single_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svqdmulh_single_s32_x211svint32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svqdmulh_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s32_x2,,,)(zdn, zm);
@@ -97,25 +103,27 @@ svint32x2_t test_svqdmulh_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_str
 
 // CHECK-LABEL: @test_svqdmulh_single_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svqdmulh_single_s64_x211svint64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svqdmulh_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s64_x2,,,)(zdn, zm);
@@ -125,37 +133,35 @@ svint64x2_t test_svqdmulh_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_str
 
 // CHECK-LABEL: @test_svqdmulh_single_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svqdmulh_single_s8_x410svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svqdmulh_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s8_x4,,,)(zdn, zm);
@@ -163,37 +169,35 @@ svint8x4_t test_svqdmulh_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streami
 
 // CHECK-LABEL: @test_svqdmulh_single_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svqdmulh_single_s16_x411svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svqdmulh_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s16_x4,,,)(zdn, zm);
@@ -201,37 +205,35 @@ svint16x4_t test_svqdmulh_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_str
 
 // CHECK-LABEL: @test_svqdmulh_single_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svqdmulh_single_s32_x411svint32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svqdmulh_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s32_x4,,,)(zdn, zm);
@@ -239,37 +241,35 @@ svint32x4_t test_svqdmulh_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_str
 
 // CHECK-LABEL: @test_svqdmulh_single_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z27test_svqdmulh_single_s64_x411svint64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.single.vgx4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svqdmulh_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_single_s64_x4,,,)(zdn, zm);
@@ -279,29 +279,27 @@ svint64x4_t test_svqdmulh_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_str
 
 // CHECK-LABEL: @test_svqdmulh_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svqdmulh_s8_x210svint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svqdmulh_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s8_x2,,,)(zdn, zm);
@@ -309,29 +307,27 @@ svint8x2_t test_svqdmulh_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svqdmulh_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svqdmulh_s16_x211svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svqdmulh_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s16_x2,,,)(zdn, zm);
@@ -339,29 +335,27 @@ svint16x2_t test_svqdmulh_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svqdmulh_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svqdmulh_s32_x211svint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svqdmulh_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s32_x2,,,)(zdn, zm);
@@ -369,29 +363,27 @@ svint32x2_t test_svqdmulh_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svqdmulh_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svqdmulh_s64_x211svint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svqdmulh_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s64_x2,,,)(zdn, zm);
@@ -401,45 +393,35 @@ svint64x2_t test_svqdmulh_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svqdmulh_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svqdmulh_s8_x410svint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svqdmulh_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s8_x4,,,)(zdn, zm);
@@ -447,45 +429,35 @@ svint8x4_t test_svqdmulh_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streaming {
 
 // CHECK-LABEL: @test_svqdmulh_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svqdmulh_s16_x411svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svqdmulh_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s16_x4,,,)(zdn, zm);
@@ -493,45 +465,35 @@ svint16x4_t test_svqdmulh_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svqdmulh_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svqdmulh_s32_x411svint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svqdmulh_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s32_x4,,,)(zdn, zm);
@@ -539,45 +501,35 @@ svint32x4_t test_svqdmulh_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svqdmulh_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svqdmulh_s64_x411svint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sqdmulh.vgx4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svqdmulh_s64_x4(svint64x4_t zdn, svint64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svqdmulh,_s64_x4,,,)(zdn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sub.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sub.c
index 6b9d5394ef8e49..aca66e0d78d667 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sub.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_sub.c
@@ -24,16 +24,12 @@
 // x2
 // CHECK-LABEL: @test_svsub_write_single2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single2_s32j11svint32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single2_s32(uint32_t slice_base, svint32x2_t zn, svint32_t zm) __arm_streaming __arm_inout("za") {
@@ -42,16 +38,12 @@ void test_svsub_write_single2_s32(uint32_t slice_base, svint32x2_t zn, svint32_t
 
 // CHECK-LABEL: @test_svsub_write_single2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single2_u32j12svuint32x2_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32_t zm) __arm_streaming __arm_inout("za") {
@@ -60,16 +52,12 @@ void test_svsub_write_single2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32
 
 // CHECK-LABEL: @test_svsub_write_single2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single2_s64j11svint64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single2_s64(uint32_t slice_base, svint64x2_t zn, svint64_t zm) __arm_streaming __arm_inout("za") {
@@ -78,16 +66,12 @@ void test_svsub_write_single2_s64(uint32_t slice_base, svint64x2_t zn, svint64_t
 
 // CHECK-LABEL: @test_svsub_write_single2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single2_u64j12svuint64x2_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64_t zm) __arm_streaming __arm_inout("za") {
@@ -98,20 +82,12 @@ void test_svsub_write_single2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64
 
 // CHECK-LABEL: @test_svsub_write_single4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single4_s32j11svint32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single4_s32(uint32_t slice_base, svint32x4_t zn, svint32_t zm) __arm_streaming __arm_inout("za") {
@@ -120,20 +96,12 @@ void test_svsub_write_single4_s32(uint32_t slice_base, svint32x4_t zn, svint32_t
 
 // CHECK-LABEL: @test_svsub_write_single4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single4_u32j12svuint32x4_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32_t zm) __arm_streaming __arm_inout("za") {
@@ -142,20 +110,12 @@ void test_svsub_write_single4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32
 
 // CHECK-LABEL: @test_svsub_write_single4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single4_s64j11svint64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single4_s64(uint32_t slice_base, svint64x4_t zn, svint64_t zm) __arm_streaming __arm_inout("za") {
@@ -164,20 +124,12 @@ void test_svsub_write_single4_s64(uint32_t slice_base, svint64x4_t zn, svint64_t
 
 // CHECK-LABEL: @test_svsub_write_single4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svsub_write_single4_u64j12svuint64x4_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.single.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_single4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64_t zm) __arm_streaming __arm_inout("za") {
@@ -192,20 +144,12 @@ void test_svsub_write_single4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64
 
 // CHECK-LABEL: @test_svsub_write_multi2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi2_s32j11svint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi2_s32(uint32_t slice_base, svint32x2_t zn, svint32x2_t zm) __arm_streaming __arm_inout("za") {
@@ -214,20 +158,12 @@ void test_svsub_write_multi2_s32(uint32_t slice_base, svint32x2_t zn, svint32x2_
 
 // CHECK-LABEL: @test_svsub_write_multi2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi2_u32j12svuint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32x2_t zm) __arm_streaming __arm_inout("za") {
@@ -236,20 +172,12 @@ void test_svsub_write_multi2_u32(uint32_t slice_base, svuint32x2_t zn, svuint32x
 
 // CHECK-LABEL: @test_svsub_write_multi2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi2_s64j11svint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi2_s64(uint32_t slice_base, svint64x2_t zn, svint64x2_t zm) __arm_streaming __arm_inout("za") {
@@ -258,20 +186,12 @@ void test_svsub_write_multi2_s64(uint32_t slice_base, svint64x2_t zn, svint64x2_
 
 // CHECK-LABEL: @test_svsub_write_multi2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi2_u64j12svuint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64x2_t zm) __arm_streaming __arm_inout("za") {
@@ -282,28 +202,12 @@ void test_svsub_write_multi2_u64(uint32_t slice_base, svuint64x2_t zn, svuint64x
 
 // CHECK-LABEL: @test_svsub_write_multi4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi4_s32j11svint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi4_s32(uint32_t slice_base, svint32x4_t zn, svint32x4_t zm) __arm_streaming __arm_inout("za") {
@@ -312,28 +216,12 @@ void test_svsub_write_multi4_s32(uint32_t slice_base, svint32x4_t zn, svint32x4_
 
 // CHECK-LABEL: @test_svsub_write_multi4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi4_u32j12svuint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32x4_t zm) __arm_streaming __arm_inout("za") {
@@ -342,28 +230,12 @@ void test_svsub_write_multi4_u32(uint32_t slice_base, svuint32x4_t zn, svuint32x
 
 // CHECK-LABEL: @test_svsub_write_multi4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi4_s64j11svint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi4_s64(uint32_t slice_base, svint64x4_t zn, svint64x4_t zm) __arm_streaming __arm_inout("za") {
@@ -372,28 +244,12 @@ void test_svsub_write_multi4_s64(uint32_t slice_base, svint64x4_t zn, svint64x4_
 
 // CHECK-LABEL: @test_svsub_write_multi4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svsub_write_multi4_u64j12svuint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.write.za.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_write_multi4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64x4_t zm) __arm_streaming __arm_inout("za") {
@@ -408,16 +264,12 @@ void test_svsub_write_multi4_u64(uint32_t slice_base, svuint64x4_t zn, svuint64x
 
 // CHECK-LABEL: @test_svsub_za32_vg1x2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za32_vg1x2_f32j13svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za32_vg1x2_f32(uint32_t slice_base, svfloat32x2_t zn) __arm_streaming __arm_inout("za") {
@@ -426,16 +278,12 @@ void test_svsub_za32_vg1x2_f32(uint32_t slice_base, svfloat32x2_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svsub_za32_vg1x2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za32_vg1x2_s32j11svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za32_vg1x2_s32(uint32_t slice_base, svint32x2_t zn) __arm_streaming __arm_inout("za") {
@@ -444,16 +292,12 @@ void test_svsub_za32_vg1x2_s32(uint32_t slice_base, svint32x2_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svsub_za32_vg1x2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za32_vg1x2_u32j12svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x2.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za32_vg1x2_u32(uint32_t slice_base, svuint32x2_t zn) __arm_streaming __arm_inout("za") {
@@ -462,16 +306,12 @@ void test_svsub_za32_vg1x2_u32(uint32_t slice_base, svuint32x2_t zn) __arm_strea
 
 // CHECK-LABEL: @test_svsub_za64_vg1x2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za64_vg1x2_f64j13svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za64_vg1x2_f64(uint32_t slice_base, svfloat64x2_t zn) __arm_streaming __arm_inout("za") {
@@ -480,16 +320,12 @@ void test_svsub_za64_vg1x2_f64(uint32_t slice_base, svfloat64x2_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svsub_za64_vg1x2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za64_vg1x2_s64j11svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za64_vg1x2_s64(uint32_t slice_base, svint64x2_t zn) __arm_streaming __arm_inout("za") {
@@ -498,16 +334,12 @@ void test_svsub_za64_vg1x2_s64(uint32_t slice_base, svint64x2_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svsub_za64_vg1x2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za64_vg1x2_u64j12svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x2.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za64_vg1x2_u64(uint32_t slice_base, svuint64x2_t zn) __arm_streaming __arm_inout("za") {
@@ -518,20 +350,12 @@ void test_svsub_za64_vg1x2_u64(uint32_t slice_base, svuint64x2_t zn) __arm_strea
 
 // CHECK-LABEL: @test_svsub_za32_vg1x4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za32_vg1x4_f32j13svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4f32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za32_vg1x4_f32(uint32_t slice_base, svfloat32x4_t zn) __arm_streaming __arm_inout("za") {
@@ -540,20 +364,12 @@ void test_svsub_za32_vg1x4_f32(uint32_t slice_base, svfloat32x4_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svsub_za32_vg1x4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za32_vg1x4_s32j11svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za32_vg1x4_s32(uint32_t slice_base, svint32x4_t zn) __arm_streaming __arm_inout("za") {
@@ -562,20 +378,12 @@ void test_svsub_za32_vg1x4_s32(uint32_t slice_base, svint32x4_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svsub_za32_vg1x4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za32_vg1x4_u32j12svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za32.vg1x4.nxv4i32(i32 [[SLICE_BASE:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za32_vg1x4_u32(uint32_t slice_base, svuint32x4_t zn) __arm_streaming __arm_inout("za") {
@@ -584,20 +392,12 @@ void test_svsub_za32_vg1x4_u32(uint32_t slice_base, svuint32x4_t zn) __arm_strea
 
 // CHECK-LABEL: @test_svsub_za64_vg1x4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za64_vg1x4_f64j13svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2f64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za64_vg1x4_f64(uint32_t slice_base, svfloat64x4_t zn) __arm_streaming __arm_inout("za") {
@@ -606,20 +406,12 @@ void test_svsub_za64_vg1x4_f64(uint32_t slice_base, svfloat64x4_t zn) __arm_stre
 
 // CHECK-LABEL: @test_svsub_za64_vg1x4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za64_vg1x4_s64j11svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za64_vg1x4_s64(uint32_t slice_base, svint64x4_t zn) __arm_streaming __arm_inout("za") {
@@ -628,20 +420,12 @@ void test_svsub_za64_vg1x4_s64(uint32_t slice_base, svint64x4_t zn) __arm_stream
 
 // CHECK-LABEL: @test_svsub_za64_vg1x4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svsub_za64_vg1x4_u64j12svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za64.vg1x4.nxv2i64(i32 [[SLICE_BASE:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsub_za64_vg1x4_u64(uint32_t slice_base, svuint64x4_t zn) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx2.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx2.c
index 0d3e06f3d3d09c..fa66c4ff190142 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx2.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx2.c
@@ -19,21 +19,27 @@
 
 // CHECK-LABEL: @test_svunpk_s16_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_s16_x2u10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svunpk_s16_x2(svint8_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_s16,_s8_x2)(zn);
@@ -41,21 +47,27 @@ svint16x2_t test_svunpk_s16_x2(svint8_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_u16_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_u16_x2u11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svunpk_u16_x2(svuint8_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_u16,_u8_x2)(zn);
@@ -63,21 +75,27 @@ svuint16x2_t test_svunpk_u16_x2(svuint8_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_s32_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_s32_x2u11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svunpk_s32_x2(svint16_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_s32,_s16_x2)(zn);
@@ -85,21 +103,27 @@ svint32x2_t test_svunpk_s32_x2(svint16_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_u32_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_u32_x2u12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svunpk_u32_x2(svuint16_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_u32,_u16_x2)(zn);
@@ -107,21 +131,27 @@ svuint32x2_t test_svunpk_u32_x2(svuint16_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_s64_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_s64_x2u11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svunpk_s64_x2(svint32_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_s64,_s32_x2)(zn);
@@ -129,21 +159,27 @@ svint64x2_t test_svunpk_s64_x2(svint32_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_u64_x2(
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_u64_x2u12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svunpk_u64_x2(svuint32_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_u64,_u32_x2)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx4.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx4.c
index e7c57e986bca4c..61718f0984ef31 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx4.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx4.c
@@ -19,33 +19,35 @@
 
 // CHECK-LABEL: @test_svunpk_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x4.nxv8i16(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 3
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP10]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x4.nxv8i16(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_s16_x410svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x4.nxv8i16(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 3
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP10]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x4.nxv8i16(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svunpk_s16_x4(svint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_s16,_s8_x4)(zn);
@@ -53,33 +55,35 @@ svint16x4_t test_svunpk_s16_x4(svint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x4.nxv8i16(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 3
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP10]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x4.nxv8i16(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_u16_x411svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x4.nxv8i16(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 3
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP10]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x4.nxv8i16(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svunpk_u16_x4(svuint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_u16,_u8_x4)(zn);
@@ -87,33 +91,35 @@ svuint16x4_t test_svunpk_u16_x4(svuint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x4.nxv4i32(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 3
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP10]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x4.nxv4i32(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_s32_x411svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x4.nxv4i32(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 3
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP10]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x4.nxv4i32(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svunpk_s32_x4(svint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_s32,_s16_x4)(zn);
@@ -121,33 +127,35 @@ svint32x4_t test_svunpk_s32_x4(svint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x4.nxv4i32(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 3
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP10]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x4.nxv4i32(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_u32_x412svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x4.nxv4i32(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 3
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP10]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x4.nxv4i32(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svunpk_u32_x4(svuint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_u32,_u16_x4)(zn);
@@ -155,33 +163,35 @@ svuint32x4_t test_svunpk_u32_x4(svuint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x4.nxv2i64(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 3
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP10]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x4.nxv2i64(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_s64_x411svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x4.nxv2i64(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 3
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP10]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x4.nxv2i64(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svunpk_s64_x4(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_s64,_s32_x4)(zn);
@@ -189,33 +199,35 @@ svint64x4_t test_svunpk_s64_x4(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svunpk_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x4.nxv2i64(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 3
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP10]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x4.nxv2i64(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svunpk_u64_x412svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x4.nxv2i64(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 3
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP10]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x4.nxv2i64(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svunpk_u64_x4(svuint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svunpk_u64,_u32_x4)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vdot.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vdot.c
index dd8b9fa0755b4a..d9445ef03b8c1d 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vdot.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vdot.c
@@ -16,16 +16,12 @@
 
 // CHECK-LABEL: @test_svvdot_lane_za32_bf16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svvdot_lane_za32_bf16_vg1x2j14svbfloat16x2_tu14__SVBfloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8bf16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za32_bf16_vg1x2(uint32_t slice_base, svbfloat16x2_t zn, svbfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -34,16 +30,12 @@ void test_svvdot_lane_za32_bf16_vg1x2(uint32_t slice_base, svbfloat16x2_t zn, sv
 
 // CHECK-LABEL: @test_svvdot_lane_za32_f16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svvdot_lane_za32_f16_vg1x2j13svfloat16x2_tu13__SVFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.fvdot.lane.za32.vg1x2.nxv8f16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za32_f16_vg1x2(uint32_t slice_base, svfloat16x2_t zn, svfloat16_t zm) __arm_streaming __arm_inout("za") {
@@ -52,16 +44,12 @@ void test_svvdot_lane_za32_f16_vg1x2(uint32_t slice_base, svfloat16x2_t zn, svfl
 
 // CHECK-LABEL: @test_svvdot_lane_za32_s16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svvdot_lane_za32_s16_vg1x2j11svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za32_s16_vg1x2(uint32_t slice_base, svint16x2_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -70,16 +58,12 @@ void test_svvdot_lane_za32_s16_vg1x2(uint32_t slice_base, svint16x2_t zn, svint1
 
 // CHECK-LABEL: @test_svvdot_lane_za32_u16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svvdot_lane_za32_u16_vg1x2j12svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x2.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za32_u16_vg1x2(uint32_t slice_base, svuint16x2_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -88,20 +72,12 @@ void test_svvdot_lane_za32_u16_vg1x2(uint32_t slice_base, svuint16x2_t zn, svuin
 
 // CHECK-LABEL: @test_svvdot_lane_za32_s8_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svvdot_lane_za32_s8_vg1x4j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za32_s8_vg1x4(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -110,20 +86,12 @@ void test_svvdot_lane_za32_s8_vg1x4(uint32_t slice_base, svint8x4_t zn, svint8_t
 
 // CHECK-LABEL: @test_svvdot_lane_za32_u8_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svvdot_lane_za32_u8_vg1x4j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za32_u8_vg1x4(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {
@@ -132,20 +100,12 @@ void test_svvdot_lane_za32_u8_vg1x4(uint32_t slice_base, svuint8x4_t zn, svuint8
 
 // CHECK-LABEL: @test_svvdot_lane_za64_s16_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svvdot_lane_za64_s16_vg1x4j11svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.svdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za64_s16_vg1x4(uint32_t slice_base, svint16x4_t zn, svint16_t zm) __arm_streaming __arm_inout("za") {
@@ -154,20 +114,12 @@ void test_svvdot_lane_za64_s16_vg1x4(uint32_t slice_base, svint16x4_t zn, svint1
 
 // CHECK-LABEL: @test_svvdot_lane_za64_u16_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z31test_svvdot_lane_za64_u16_vg1x4j12svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.uvdot.lane.za64.vg1x4.nxv8i16(i32 [[SLICE_BASE:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]], i32 1)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svvdot_lane_za64_u16_vg1x4(uint32_t slice_base, svuint16x4_t zn, svuint16_t zm) __arm_streaming __arm_inout("za") {
@@ -177,20 +129,12 @@ void test_svvdot_lane_za64_u16_vg1x4(uint32_t slice_base, svuint16x4_t zn, svuin
 
 // CHECK-LABEL: @test_svsuvdot_lane_za32_s8_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.suvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.suvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svsuvdot_lane_za32_s8_vg1x4j10svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.suvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.suvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svsuvdot_lane_za32_s8_vg1x4(uint32_t slice_base, svint8x4_t zn, svint8_t zm) __arm_streaming __arm_inout("za") {
@@ -200,20 +144,12 @@ void test_svsuvdot_lane_za32_s8_vg1x4(uint32_t slice_base, svint8x4_t zn, svint8
 
 // CHECK-LABEL: @test_svusvdot_lane_za32_u8_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.usvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z32test_svusvdot_lane_za32_u8_vg1x4j11svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.usvdot.lane.za32.vg1x4.nxv16i8(i32 [[SLICE_BASE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svusvdot_lane_za32_u8_vg1x4(uint32_t slice_base, svuint8x4_t zn, svuint8_t zm) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_add.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_add.c
index 425fa55503174c..c118a7192c6ca8 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_add.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_add.c
@@ -25,25 +25,27 @@
 
 // CHECK-LABEL: @test_svadd_vector_single2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_vector_single2_s810svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svadd_vector_single2_s8(svint8x2_t zn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s8_x2,,,)(zn, zm);
@@ -51,25 +53,27 @@ svint8x2_t test_svadd_vector_single2_s8(svint8x2_t zn, svint8_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svadd_vector_single2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_vector_single2_u811svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svadd_vector_single2_u8(svuint8x2_t zn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u8_x2,,,)(zn, zm);
@@ -77,25 +81,27 @@ svuint8x2_t test_svadd_vector_single2_u8(svuint8x2_t zn, svuint8_t zm) __arm_str
 
 // CHECK-LABEL: @test_svadd_vector_single2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single2_s1611svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svadd_vector_single2_s16(svint16x2_t zn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s16_x2,,,)(zn, zm);
@@ -103,25 +109,27 @@ svint16x2_t test_svadd_vector_single2_s16(svint16x2_t zn, svint16_t zm) __arm_st
 
 // CHECK-LABEL: @test_svadd_vector_single2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single2_u1612svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svadd_vector_single2_u16(svuint16x2_t zn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u16_x2,,,)(zn, zm);
@@ -129,25 +137,27 @@ svuint16x2_t test_svadd_vector_single2_u16(svuint16x2_t zn, svuint16_t zm) __arm
 
 // CHECK-LABEL: @test_svadd_vector_single2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single2_s3211svint32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svadd_vector_single2_s32(svint32x2_t zn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s32_x2,,,)(zn, zm);
@@ -155,25 +165,27 @@ svint32x2_t test_svadd_vector_single2_s32(svint32x2_t zn, svint32_t zm) __arm_st
 
 // CHECK-LABEL: @test_svadd_vector_single2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single2_u3212svuint32x2_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svadd_vector_single2_u32(svuint32x2_t zn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u32_x2,,,)(zn, zm);
@@ -181,25 +193,27 @@ svuint32x2_t test_svadd_vector_single2_u32(svuint32x2_t zn, svuint32_t zm) __arm
 
 // CHECK-LABEL: @test_svadd_vector_single2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single2_s6411svint64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svadd_vector_single2_s64(svint64x2_t zn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s64_x2,,,)(zn, zm);
@@ -207,25 +221,27 @@ svint64x2_t test_svadd_vector_single2_s64(svint64x2_t zn, svint64_t zm) __arm_st
 
 // CHECK-LABEL: @test_svadd_vector_single2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single2_u6412svuint64x2_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svadd_vector_single2_u64(svuint64x2_t zn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u64_x2,,,)(zn, zm);
@@ -236,37 +252,35 @@ svuint64x2_t test_svadd_vector_single2_u64(svuint64x2_t zn, svuint64_t zm) __arm
 
 // CHECK-LABEL: @test_svadd_vector_single4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_vector_single4_s810svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svadd_vector_single4_s8(svint8x4_t zn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s8_x4,,,)(zn, zm);
@@ -274,37 +288,35 @@ svint8x4_t test_svadd_vector_single4_s8(svint8x4_t zn, svint8_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svadd_vector_single4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z28test_svadd_vector_single4_u811svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.add.single.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svadd_vector_single4_u8(svuint8x4_t zn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u8_x4,,,)(zn, zm);
@@ -312,37 +324,35 @@ svuint8x4_t test_svadd_vector_single4_u8(svuint8x4_t zn, svuint8_t zm) __arm_str
 
 // CHECK-LABEL: @test_svadd_vector_single4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single4_s1611svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svadd_vector_single4_s16(svint16x4_t zn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s16_x4,,,)(zn, zm);
@@ -350,37 +360,35 @@ svint16x4_t test_svadd_vector_single4_s16(svint16x4_t zn, svint16_t zm) __arm_st
 
 // CHECK-LABEL: @test_svadd_vector_single4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single4_u1612svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.add.single.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svadd_vector_single4_u16(svuint16x4_t zn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u16_x4,,,)(zn, zm);
@@ -388,37 +396,35 @@ svuint16x4_t test_svadd_vector_single4_u16(svuint16x4_t zn, svuint16_t zm) __arm
 
 // CHECK-LABEL: @test_svadd_vector_single4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single4_s3211svint32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svadd_vector_single4_s32(svint32x4_t zn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s32_x4,,,)(zn, zm);
@@ -426,37 +432,35 @@ svint32x4_t test_svadd_vector_single4_s32(svint32x4_t zn, svint32_t zm) __arm_st
 
 // CHECK-LABEL: @test_svadd_vector_single4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single4_u3212svuint32x4_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.add.single.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svadd_vector_single4_u32(svuint32x4_t zn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u32_x4,,,)(zn, zm);
@@ -464,37 +468,35 @@ svuint32x4_t test_svadd_vector_single4_u32(svuint32x4_t zn, svuint32_t zm) __arm
 
 // CHECK-LABEL: @test_svadd_vector_single4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single4_s6411svint64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svadd_vector_single4_s64(svint64x4_t zn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_s64_x4,,,)(zn, zm);
@@ -502,37 +504,35 @@ svint64x4_t test_svadd_vector_single4_s64(svint64x4_t zn, svint64_t zm) __arm_st
 
 // CHECK-LABEL: @test_svadd_vector_single4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z29test_svadd_vector_single4_u6412svuint64x4_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.add.single.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svadd_vector_single4_u64(svuint64x4_t zn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svadd,_single_u64_x4,,,)(zn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_qrshr.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_qrshr.c
index 9a6fc712143541..3e47a3ecc17c0b 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_qrshr.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_qrshr.c
@@ -20,17 +20,13 @@
 
 // CHECK-LABEL: @test_svsqrshr_u16_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svsqrshr_u16_u32_x412svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_svsqrshr_u16_u32_x4(svuint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshr,_n,_u16,_u32_x2,)(zn, 16);
@@ -38,17 +34,13 @@ svuint16_t test_svsqrshr_u16_u32_x4(svuint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshr_s16_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svsqrshr_s16_s32_x411svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svint16_t test_svsqrshr_s16_s32_x4(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshr,_n,_s16,_s32_x2,)(zn, 16);
@@ -56,21 +48,13 @@ svint16_t test_svsqrshr_s16_s32_x4(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshr_u8_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshr.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshr.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svsqrshr_u8_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshr.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshr.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_svsqrshr_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshr,_n,_u8,_u32_x4,)(zn, 8);
@@ -78,21 +62,13 @@ svuint8_t test_svsqrshr_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshr_s8_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshr.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshr.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svsqrshr_s8_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshr.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshr.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svint8_t test_svsqrshr_s8_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshr,_n,_s8,_s32_x4,)(zn, 8);
@@ -100,21 +76,13 @@ svint8_t test_svsqrshr_s8_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshr_u16_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svsqrshr_u16_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshr.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_svsqrshr_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshr,_n,_u16,_u64_x4,)(zn, 16);
@@ -122,21 +90,13 @@ svuint16_t test_svsqrshr_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshr_s16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svsqrshr_s16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshr.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svint16_t test_svsqrshr_s16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshr,_n,_s16,_s64_x4,)(zn, 16);
@@ -146,21 +106,13 @@ svint16_t test_svsqrshr_s16_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshrn_u8_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svsqrshrn_u8_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.uqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_svsqrshrn_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshrn,_n,_u8,_u32_x4,)(zn, 8);
@@ -168,21 +120,13 @@ svuint8_t test_svsqrshrn_u8_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshrn_s8_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svsqrshrn_s8_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrn.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svint8_t test_svsqrshrn_s8_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshrn,_n,_s8,_s32_x4,)(zn, 8);
@@ -190,21 +134,13 @@ svint8_t test_svsqrshrn_s8_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshrn_u16_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svsqrshrn_u16_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.uqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_svsqrshrn_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshrn,_n,_u16,_u64_x4,)(zn, 16);
@@ -212,21 +148,13 @@ svuint16_t test_svsqrshrn_u16_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshrn_s16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svsqrshrn_s16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrn.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svint16_t test_svsqrshrn_s16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshrn,_n,_s16,_s64_x4,)(zn, 16);
@@ -236,17 +164,13 @@ svint16_t test_svsqrshrn_s16_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsvqrshru_u16_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svsvqrshru_u16_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_svsvqrshru_u16_s32_x2(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshru,_n,_u16,_s32_x2,)(zn, 16);
@@ -254,21 +178,13 @@ svuint16_t test_svsvqrshru_u16_s32_x2(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshru_u8_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshru.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshru.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svsqrshru_u8_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshru.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshru.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_svsqrshru_u8_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshru,_n,_u8,_s32_x4,)(zn, 8);
@@ -276,21 +192,13 @@ svuint8_t test_svsqrshru_u8_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshru_u16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svsqrshru_u16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshru.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 16)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_svsqrshru_u16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshru,_n,_u16,_s64_x4,)(zn, 16);
@@ -300,21 +208,13 @@ svuint16_t test_svsqrshru_u16_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshrun_u8_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrun.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 32)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrun.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 32)
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svsqrshrun_u8_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrun.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i32 32)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.sqrshrun.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]], i32 32)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 svuint8_t test_svsqrshrun_u8_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshrun,_n,_u8,_s32_x4,)(zn, 32);
@@ -322,21 +222,13 @@ svuint8_t test_svsqrshrun_u8_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svsqrshrun_u16_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrun.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 64)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrun.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 64)
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svsqrshrun_u16_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrun.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i32 64)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.sqrshrun.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]], i32 64)
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
 svuint16_t test_svsqrshrun_u16_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svqrshrun,_n,_u16,_s64_x4,)(zn, 64);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_rshl.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_rshl.c
index f69a94c15b8580..87160444e3c0d1 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_rshl.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_rshl.c
@@ -19,25 +19,27 @@
 
 // CHECK-LABEL: @test_svrshl_single_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_single_s8_x210svint8x2_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svrshl_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s8_x2,,,)(zdn, zm);
@@ -45,25 +47,27 @@ svint8x2_t test_svrshl_single_s8_x2(svint8x2_t zdn, svint8_t zm) __arm_streaming
 
 // CHECK-LABEL: @test_svrshl_single_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_s16_x211svint16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svrshl_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s16_x2,,,)(zdn, zm);
@@ -71,25 +75,27 @@ svint16x2_t test_svrshl_single_s16_x2(svint16x2_t zdn, svint16_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_single_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_s32_x211svint32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svrshl_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s32_x2,,,)(zdn, zm);
@@ -97,25 +103,27 @@ svint32x2_t test_svrshl_single_s32_x2(svint32x2_t zdn, svint32_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_single_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_s64_x211svint64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svrshl_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s64_x2,,,)(zdn, zm);
@@ -123,25 +131,27 @@ svint64x2_t test_svrshl_single_s64_x2(svint64x2_t zdn, svint64_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_single_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_single_u8_x211svuint8x2_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svrshl_single_u8_x2(svuint8x2_t zdn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u8_x2,,,)(zdn, zm);
@@ -149,25 +159,27 @@ svuint8x2_t test_svrshl_single_u8_x2(svuint8x2_t zdn, svuint8_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svrshl_single_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_u16_x212svuint16x2_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svrshl_single_u16_x2(svuint16x2_t zdn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u16_x2,,,)(zdn, zm);
@@ -175,25 +187,27 @@ svuint16x2_t test_svrshl_single_u16_x2(svuint16x2_t zdn, svuint16_t zm) __arm_st
 
 // CHECK-LABEL: @test_svrshl_single_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_u32_x212svuint32x2_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svrshl_single_u32_x2(svuint32x2_t zdn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u32_x2,,,)(zdn, zm);
@@ -201,25 +215,27 @@ svuint32x2_t test_svrshl_single_u32_x2(svuint32x2_t zdn, svuint32_t zm) __arm_st
 
 // CHECK-LABEL: @test_svrshl_single_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_u64_x212svuint64x2_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svrshl_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u64_x2,,,)(zdn, zm);
@@ -229,37 +245,35 @@ svuint64x2_t test_svrshl_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_st
 
 // CHECK-LABEL: @test_svrshl_single_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_single_s8_x410svint8x4_tu10__SVInt8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svrshl_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s8_x4,,,)(zdn, zm);
@@ -267,37 +281,35 @@ svint8x4_t test_svrshl_single_s8_x4(svint8x4_t zdn, svint8_t zm) __arm_streaming
 
 // CHECK-LABEL: @test_svrshl_single_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_s16_x411svint16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svrshl_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s16_x4,,,)(zdn, zm);
@@ -305,37 +317,35 @@ svint16x4_t test_svrshl_single_s16_x4(svint16x4_t zdn, svint16_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_single_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_s32_x411svint32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svrshl_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s32_x4,,,)(zdn, zm);
@@ -343,37 +353,35 @@ svint32x4_t test_svrshl_single_s32_x4(svint32x4_t zdn, svint32_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_single_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_s64_x411svint64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svrshl_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_s64_x4,,,)(zdn, zm);
@@ -381,37 +389,35 @@ svint64x4_t test_svrshl_single_s64_x4(svint64x4_t zdn, svint64_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_single_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_single_u8_x411svuint8x4_tu11__SVUint8_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.single.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svrshl_single_u8_x4(svuint8x4_t zdn, svuint8_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u8_x4,,,)(zdn, zm);
@@ -419,37 +425,35 @@ svuint8x4_t test_svrshl_single_u8_x4(svuint8x4_t zdn, svuint8_t zm) __arm_stream
 
 // CHECK-LABEL: @test_svrshl_single_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_u16_x412svuint16x4_tu12__SVUint16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.single.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svrshl_single_u16_x4(svuint16x4_t zdn, svuint16_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u16_x4,,,)(zdn, zm);
@@ -457,37 +461,35 @@ svuint16x4_t test_svrshl_single_u16_x4(svuint16x4_t zdn, svuint16_t zm) __arm_st
 
 // CHECK-LABEL: @test_svrshl_single_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_u32_x412svuint32x4_tu12__SVUint32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.single.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svrshl_single_u32_x4(svuint32x4_t zdn, svuint32_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u32_x4,,,)(zdn, zm);
@@ -495,37 +497,35 @@ svuint32x4_t test_svrshl_single_u32_x4(svuint32x4_t zdn, svuint32_t zm) __arm_st
 
 // CHECK-LABEL: @test_svrshl_single_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z25test_svrshl_single_u64_x412svuint64x4_tu12__SVUint64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[ZM:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.single.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svrshl_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_single_u64_x4,,,)(zdn, zm);
@@ -535,29 +535,27 @@ svuint64x4_t test_svrshl_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_st
 
 // CHECK-LABEL: @test_svrshl_multi_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svrshl_multi_s8_x210svint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svrshl_multi_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s8_x2,,,)(zdn, zm);
@@ -565,29 +563,27 @@ svint8x2_t test_svrshl_multi_s8_x2(svint8x2_t zdn, svint8x2_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svrshl_multi_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_s16_x211svint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svrshl_multi_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s16_x2,,,)(zdn, zm);
@@ -595,29 +591,27 @@ svint16x2_t test_svrshl_multi_s16_x2(svint16x2_t zdn, svint16x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svrshl_multi_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_s32_x211svint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svrshl_multi_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s32_x2,,,)(zdn, zm);
@@ -625,29 +619,27 @@ svint32x2_t test_svrshl_multi_s32_x2(svint32x2_t zdn, svint32x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svrshl_multi_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_s64_x211svint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svrshl_multi_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s64_x2,,,)(zdn, zm);
@@ -655,29 +647,27 @@ svint64x2_t test_svrshl_multi_s64_x2(svint64x2_t zdn, svint64x2_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svrshl_multi_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svrshl_multi_u8_x211svuint8x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x2.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svrshl_multi_u8_x2(svuint8x2_t zdn, svuint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u8_x2,,,)(zdn, zm);
@@ -685,29 +675,27 @@ svuint8x2_t test_svrshl_multi_u8_x2(svuint8x2_t zdn, svuint8x2_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_multi_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_u16_x212svuint16x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x2.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svrshl_multi_u16_x2(svuint16x2_t zdn, svuint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u16_x2,,,)(zdn, zm);
@@ -715,29 +703,27 @@ svuint16x2_t test_svrshl_multi_u16_x2(svuint16x2_t zdn, svuint16x2_t zm) __arm_s
 
 // CHECK-LABEL: @test_svrshl_multi_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_u32_x212svuint32x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x2.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svrshl_multi_u32_x2(svuint32x2_t zdn, svuint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u32_x2,,,)(zdn, zm);
@@ -745,29 +731,27 @@ svuint32x2_t test_svrshl_multi_u32_x2(svuint32x2_t zdn, svuint32x2_t zm) __arm_s
 
 // CHECK-LABEL: @test_svrshl_multi_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_u64_x212svuint64x2_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x2.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svrshl_multi_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u64_x2,,,)(zdn, zm);
@@ -777,45 +761,35 @@ svuint64x2_t test_svrshl_multi_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_s
 
 // CHECK-LABEL: @test_svrshl_multi_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svrshl_multi_s8_x410svint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.srshl.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svrshl_multi_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s8_x4,,,)(zdn, zm);
@@ -823,45 +797,35 @@ svint8x4_t test_svrshl_multi_s8_x4(svint8x4_t zdn, svint8x4_t zm) __arm_streamin
 
 // CHECK-LABEL: @test_svrshl_multi_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_s16_x411svint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.srshl.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svrshl_multi_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s16_x4,,,)(zdn, zm);
@@ -869,45 +833,35 @@ svint16x4_t test_svrshl_multi_s16_x4(svint16x4_t zdn, svint16x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svrshl_multi_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_s32_x411svint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.srshl.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svrshl_multi_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s32_x4,,,)(zdn, zm);
@@ -915,45 +869,35 @@ svint32x4_t test_svrshl_multi_s32_x4(svint32x4_t zdn, svint32x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svrshl_multi_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_s64_x411svint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.srshl.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svrshl_multi_s64_x4(svint64x4_t zdn, svint64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_s64_x4,,,)(zdn, zm);
@@ -961,45 +905,35 @@ svint64x4_t test_svrshl_multi_s64_x4(svint64x4_t zdn, svint64x4_t zm) __arm_stre
 
 // CHECK-LABEL: @test_svrshl_multi_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z23test_svrshl_multi_u8_x411svuint8x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZDN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZM]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.urshl.x4.nxv16i8(<vscale x 16 x i8> [[ZDN_COERCE0:%.*]], <vscale x 16 x i8> [[ZDN_COERCE1:%.*]], <vscale x 16 x i8> [[ZDN_COERCE2:%.*]], <vscale x 16 x i8> [[ZDN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE2:%.*]], <vscale x 16 x i8> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svrshl_multi_u8_x4(svuint8x4_t zdn, svuint8x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u8_x4,,,)(zdn, zm);
@@ -1007,45 +941,35 @@ svuint8x4_t test_svrshl_multi_u8_x4(svuint8x4_t zdn, svuint8x4_t zm) __arm_strea
 
 // CHECK-LABEL: @test_svrshl_multi_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_u16_x412svuint16x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZDN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZM]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.urshl.x4.nxv8i16(<vscale x 8 x i16> [[ZDN_COERCE0:%.*]], <vscale x 8 x i16> [[ZDN_COERCE1:%.*]], <vscale x 8 x i16> [[ZDN_COERCE2:%.*]], <vscale x 8 x i16> [[ZDN_COERCE3:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE2:%.*]], <vscale x 8 x i16> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svrshl_multi_u16_x4(svuint16x4_t zdn, svuint16x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u16_x4,,,)(zdn, zm);
@@ -1053,45 +977,35 @@ svuint16x4_t test_svrshl_multi_u16_x4(svuint16x4_t zdn, svuint16x4_t zm) __arm_s
 
 // CHECK-LABEL: @test_svrshl_multi_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_u32_x412svuint32x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZDN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZM]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.urshl.x4.nxv4i32(<vscale x 4 x i32> [[ZDN_COERCE0:%.*]], <vscale x 4 x i32> [[ZDN_COERCE1:%.*]], <vscale x 4 x i32> [[ZDN_COERCE2:%.*]], <vscale x 4 x i32> [[ZDN_COERCE3:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE2:%.*]], <vscale x 4 x i32> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svrshl_multi_u32_x4(svuint32x4_t zdn, svuint32x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u32_x4,,,)(zdn, zm);
@@ -1099,45 +1013,35 @@ svuint32x4_t test_svrshl_multi_u32_x4(svuint32x4_t zdn, svuint32x4_t zm) __arm_s
 
 // CHECK-LABEL: @test_svrshl_multi_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z24test_svrshl_multi_u64_x412svuint64x4_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZDN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZM]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.urshl.x4.nxv2i64(<vscale x 2 x i64> [[ZDN_COERCE0:%.*]], <vscale x 2 x i64> [[ZDN_COERCE1:%.*]], <vscale x 2 x i64> [[ZDN_COERCE2:%.*]], <vscale x 2 x i64> [[ZDN_COERCE3:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE2:%.*]], <vscale x 2 x i64> [[ZM_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svrshl_multi_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svrshl,_u64_x4,,,)(zdn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx2.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx2.c
index 47d449515ce669..a95f89faf77834 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx2.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx2.c
@@ -19,29 +19,27 @@
 
 // CHECK-LABEL: @test_svsel_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svsel_s8_x2u11__SVCount_t10svint8x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svsel_s8_x2(svcount_t pn, svint8x2_t zn, svint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s8_x2)(pn, zn, zm);
@@ -49,29 +47,27 @@ svint8x2_t test_svsel_s8_x2(svcount_t pn, svint8x2_t zn, svint8x2_t zm) __arm_st
 
 // CHECK-LABEL: @test_svsel_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svsel_u8_x2u11__SVCount_t11svuint8x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZM]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x2.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svsel_u8_x2(svcount_t pn, svuint8x2_t zn, svuint8x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u8_x2)(pn, zn, zm);
@@ -81,29 +77,27 @@ svuint8x2_t test_svsel_u8_x2(svcount_t pn, svuint8x2_t zn, svuint8x2_t zm) __arm
 
 // CHECK-LABEL: @test_svsel_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_s16_x2u11__SVCount_t11svint16x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svsel_s16_x2(svcount_t pn, svint16x2_t zn, svint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s16_x2)(pn, zn, zm);
@@ -111,29 +105,27 @@ svint16x2_t test_svsel_s16_x2(svcount_t pn, svint16x2_t zn, svint16x2_t zm) __ar
 
 // CHECK-LABEL: @test_svsel_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_u16_x2u11__SVCount_t12svuint16x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x2.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZM_COERCE0:%.*]], <vscale x 8 x i16> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svsel_u16_x2(svcount_t pn, svuint16x2_t zn, svuint16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u16_x2)(pn, zn, zm);
@@ -141,29 +133,27 @@ svuint16x2_t test_svsel_u16_x2(svcount_t pn, svuint16x2_t zn, svuint16x2_t zm) _
 
 // CHECK-LABEL: @test_svsel_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x2.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x2.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_f16_x2u11__SVCount_t13svfloat16x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x2.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x2.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svsel_f16_x2(svcount_t pn, svfloat16x2_t zn, svfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_f16_x2)(pn, zn, zm);
@@ -171,29 +161,27 @@ svfloat16x2_t test_svsel_f16_x2(svcount_t pn, svfloat16x2_t zn, svfloat16x2_t zm
 
 // CHECK-LABEL: @test_svsel_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x2.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x2.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svsel_bf16_x2u11__SVCount_t14svbfloat16x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x2.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x2.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svsel_bf16_x2(svcount_t pn, svbfloat16x2_t zn, svbfloat16x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_bf16_x2)(pn, zn, zm);
@@ -203,29 +191,27 @@ svbfloat16x2_t test_svsel_bf16_x2(svcount_t pn, svbfloat16x2_t zn, svbfloat16x2_
 
 // CHECK-LABEL: @test_svsel_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_s32_x2u11__SVCount_t11svint32x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svsel_s32_x2(svcount_t pn, svint32x2_t zn, svint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s32_x2)(pn, zn, zm);
@@ -233,29 +219,27 @@ svint32x2_t test_svsel_s32_x2(svcount_t pn, svint32x2_t zn, svint32x2_t zm) __ar
 
 // CHECK-LABEL: @test_svsel_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_u32_x2u11__SVCount_t12svuint32x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x2.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZM_COERCE0:%.*]], <vscale x 4 x i32> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svsel_u32_x2(svcount_t pn, svuint32x2_t zn, svuint32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u32_x2)(pn, zn, zm);
@@ -263,29 +247,27 @@ svuint32x2_t test_svsel_u32_x2(svcount_t pn, svuint32x2_t zn, svuint32x2_t zm) _
 
 // CHECK-LABEL: @test_svsel_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x2.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x2.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_f32_x2u11__SVCount_t13svfloat32x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x2.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x2.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZM_COERCE0:%.*]], <vscale x 4 x float> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svsel_f32_x2(svcount_t pn, svfloat32x2_t zn, svfloat32x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_f32_x2)(pn, zn, zm);
@@ -295,29 +277,27 @@ svfloat32x2_t test_svsel_f32_x2(svcount_t pn, svfloat32x2_t zn, svfloat32x2_t zm
 
 // CHECK-LABEL: @test_svsel_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_s64_x2u11__SVCount_t11svint64x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svsel_s64_x2(svcount_t pn, svint64x2_t zn, svint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s64_x2)(pn, zn, zm);
@@ -325,29 +305,27 @@ svint64x2_t test_svsel_s64_x2(svcount_t pn, svint64x2_t zn, svint64x2_t zm) __ar
 
 // CHECK-LABEL: @test_svsel_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_u64_x2u11__SVCount_t12svuint64x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x2.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZM_COERCE0:%.*]], <vscale x 2 x i64> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svsel_u64_x2(svcount_t pn, svuint64x2_t zn, svuint64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u64_x2)(pn, zn, zm);
@@ -355,29 +333,27 @@ svuint64x2_t test_svsel_u64_x2(svcount_t pn, svuint64x2_t zn, svuint64x2_t zm) _
 
 // CHECK-LABEL: @test_svsel_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x2.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x2.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_f64_x2u11__SVCount_t13svfloat64x2_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x2.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x2.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZM_COERCE0:%.*]], <vscale x 2 x double> [[ZM_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svsel_f64_x2(svcount_t pn, svfloat64x2_t zn, svfloat64x2_t zm) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_f64_x2)(pn, zn, zm);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx4.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx4.c
index 88de2ec2f8ace0..997b6acf962443 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx4.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_selx4.c
@@ -19,45 +19,35 @@
 
 // CHECK-LABEL: @test_svsel_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN1_COERCE0:%.*]], <vscale x 16 x i8> [[ZN1_COERCE1:%.*]], <vscale x 16 x i8> [[ZN1_COERCE2:%.*]], <vscale x 16 x i8> [[ZN1_COERCE3:%.*]], <vscale x 16 x i8> [[ZN2_COERCE0:%.*]], <vscale x 16 x i8> [[ZN2_COERCE1:%.*]], <vscale x 16 x i8> [[ZN2_COERCE2:%.*]], <vscale x 16 x i8> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svsel_s8_x4u11__SVCount_t10svint8x4_tS0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN1_COERCE0:%.*]], <vscale x 16 x i8> [[ZN1_COERCE1:%.*]], <vscale x 16 x i8> [[ZN1_COERCE2:%.*]], <vscale x 16 x i8> [[ZN1_COERCE3:%.*]], <vscale x 16 x i8> [[ZN2_COERCE0:%.*]], <vscale x 16 x i8> [[ZN2_COERCE1:%.*]], <vscale x 16 x i8> [[ZN2_COERCE2:%.*]], <vscale x 16 x i8> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svsel_s8_x4(svcount_t pn, svint8x4_t zn1, svint8x4_t zn2) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s8_x4)(pn, zn1, zn2);
@@ -65,45 +55,35 @@ svint8x4_t test_svsel_s8_x4(svcount_t pn, svint8x4_t zn1, svint8x4_t zn2) __arm_
 
 // CHECK-LABEL: @test_svsel_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 48)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN1_COERCE0:%.*]], <vscale x 16 x i8> [[ZN1_COERCE1:%.*]], <vscale x 16 x i8> [[ZN1_COERCE2:%.*]], <vscale x 16 x i8> [[ZN1_COERCE3:%.*]], <vscale x 16 x i8> [[ZN2_COERCE0:%.*]], <vscale x 16 x i8> [[ZN2_COERCE1:%.*]], <vscale x 16 x i8> [[ZN2_COERCE2:%.*]], <vscale x 16 x i8> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svsel_u8_x4u11__SVCount_t11svuint8x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN1]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN2]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP12]], <vscale x 16 x i8> [[TMP13]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP14]], <vscale x 16 x i8> [[TMP15]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.sel.x4.nxv16i8(target("aarch64.svcount") [[PN:%.*]], <vscale x 16 x i8> [[ZN1_COERCE0:%.*]], <vscale x 16 x i8> [[ZN1_COERCE1:%.*]], <vscale x 16 x i8> [[ZN1_COERCE2:%.*]], <vscale x 16 x i8> [[ZN1_COERCE3:%.*]], <vscale x 16 x i8> [[ZN2_COERCE0:%.*]], <vscale x 16 x i8> [[ZN2_COERCE1:%.*]], <vscale x 16 x i8> [[ZN2_COERCE2:%.*]], <vscale x 16 x i8> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svsel_u8_x4(svcount_t pn, svuint8x4_t zn1, svuint8x4_t zn2, svuint8x4_t zn3, svuint8x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u8_x4)(pn, zn1, zn2);
@@ -113,45 +93,35 @@ svuint8x4_t test_svsel_u8_x4(svcount_t pn, svuint8x4_t zn1, svuint8x4_t zn2, svu
 
 // CHECK-LABEL: @test_svsel_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN1_COERCE0:%.*]], <vscale x 8 x i16> [[ZN1_COERCE1:%.*]], <vscale x 8 x i16> [[ZN1_COERCE2:%.*]], <vscale x 8 x i16> [[ZN1_COERCE3:%.*]], <vscale x 8 x i16> [[ZN2_COERCE0:%.*]], <vscale x 8 x i16> [[ZN2_COERCE1:%.*]], <vscale x 8 x i16> [[ZN2_COERCE2:%.*]], <vscale x 8 x i16> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_s16_x4u11__SVCount_t11svint16x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN1_COERCE0:%.*]], <vscale x 8 x i16> [[ZN1_COERCE1:%.*]], <vscale x 8 x i16> [[ZN1_COERCE2:%.*]], <vscale x 8 x i16> [[ZN1_COERCE3:%.*]], <vscale x 8 x i16> [[ZN2_COERCE0:%.*]], <vscale x 8 x i16> [[ZN2_COERCE1:%.*]], <vscale x 8 x i16> [[ZN2_COERCE2:%.*]], <vscale x 8 x i16> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svsel_s16_x4(svcount_t pn, svint16x4_t zn1, svint16x4_t zn2, svint16x4_t zn3, svint16x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s16_x4)(pn, zn1, zn2);
@@ -159,45 +129,35 @@ svint16x4_t test_svsel_s16_x4(svcount_t pn, svint16x4_t zn1, svint16x4_t zn2, sv
 
 // CHECK-LABEL: @test_svsel_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN1_COERCE0:%.*]], <vscale x 8 x i16> [[ZN1_COERCE1:%.*]], <vscale x 8 x i16> [[ZN1_COERCE2:%.*]], <vscale x 8 x i16> [[ZN1_COERCE3:%.*]], <vscale x 8 x i16> [[ZN2_COERCE0:%.*]], <vscale x 8 x i16> [[ZN2_COERCE1:%.*]], <vscale x 8 x i16> [[ZN2_COERCE2:%.*]], <vscale x 8 x i16> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_u16_x4u11__SVCount_t12svuint16x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN2]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sel.x4.nxv8i16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x i16> [[ZN1_COERCE0:%.*]], <vscale x 8 x i16> [[ZN1_COERCE1:%.*]], <vscale x 8 x i16> [[ZN1_COERCE2:%.*]], <vscale x 8 x i16> [[ZN1_COERCE3:%.*]], <vscale x 8 x i16> [[ZN2_COERCE0:%.*]], <vscale x 8 x i16> [[ZN2_COERCE1:%.*]], <vscale x 8 x i16> [[ZN2_COERCE2:%.*]], <vscale x 8 x i16> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svsel_u16_x4(svcount_t pn, svuint16x4_t zn1, svuint16x4_t zn2, svuint16x4_t zn3, svuint16x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u16_x4)(pn, zn1, zn2);
@@ -205,45 +165,35 @@ svuint16x4_t test_svsel_u16_x4(svcount_t pn, svuint16x4_t zn1, svuint16x4_t zn2,
 
 // CHECK-LABEL: @test_svsel_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x4.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x4.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[ZN1_COERCE0:%.*]], <vscale x 8 x half> [[ZN1_COERCE1:%.*]], <vscale x 8 x half> [[ZN1_COERCE2:%.*]], <vscale x 8 x half> [[ZN1_COERCE3:%.*]], <vscale x 8 x half> [[ZN2_COERCE0:%.*]], <vscale x 8 x half> [[ZN2_COERCE1:%.*]], <vscale x 8 x half> [[ZN2_COERCE2:%.*]], <vscale x 8 x half> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_f16_x4u11__SVCount_t13svfloat16x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN2]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x4.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.sel.x4.nxv8f16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x half> [[ZN1_COERCE0:%.*]], <vscale x 8 x half> [[ZN1_COERCE1:%.*]], <vscale x 8 x half> [[ZN1_COERCE2:%.*]], <vscale x 8 x half> [[ZN1_COERCE3:%.*]], <vscale x 8 x half> [[ZN2_COERCE0:%.*]], <vscale x 8 x half> [[ZN2_COERCE1:%.*]], <vscale x 8 x half> [[ZN2_COERCE2:%.*]], <vscale x 8 x half> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svsel_f16_x4(svcount_t pn, svfloat16x4_t zn1, svfloat16x4_t zn2, svfloat16x4_t zn3, svfloat16x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_f16_x4)(pn, zn1, zn2);
@@ -251,45 +201,35 @@ svfloat16x4_t test_svsel_f16_x4(svcount_t pn, svfloat16x4_t zn1, svfloat16x4_t z
 
 // CHECK-LABEL: @test_svsel_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x4.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x4.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svsel_bf16_x4u11__SVCount_t14svbfloat16x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN2]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x4.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.sel.x4.nxv8bf16(target("aarch64.svcount") [[PN:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN1_COERCE3:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svsel_bf16_x4(svcount_t pn, svbfloat16x4_t zn1, svbfloat16x4_t zn2, svbfloat16x4_t zn3, svbfloat16x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_bf16_x4)(pn, zn1, zn2);
@@ -299,45 +239,35 @@ svbfloat16x4_t test_svsel_bf16_x4(svcount_t pn, svbfloat16x4_t zn1, svbfloat16x4
 
 // CHECK-LABEL: @test_svsel_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN1_COERCE0:%.*]], <vscale x 4 x i32> [[ZN1_COERCE1:%.*]], <vscale x 4 x i32> [[ZN1_COERCE2:%.*]], <vscale x 4 x i32> [[ZN1_COERCE3:%.*]], <vscale x 4 x i32> [[ZN2_COERCE0:%.*]], <vscale x 4 x i32> [[ZN2_COERCE1:%.*]], <vscale x 4 x i32> [[ZN2_COERCE2:%.*]], <vscale x 4 x i32> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_s32_x4u11__SVCount_t11svint32x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN1_COERCE0:%.*]], <vscale x 4 x i32> [[ZN1_COERCE1:%.*]], <vscale x 4 x i32> [[ZN1_COERCE2:%.*]], <vscale x 4 x i32> [[ZN1_COERCE3:%.*]], <vscale x 4 x i32> [[ZN2_COERCE0:%.*]], <vscale x 4 x i32> [[ZN2_COERCE1:%.*]], <vscale x 4 x i32> [[ZN2_COERCE2:%.*]], <vscale x 4 x i32> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svsel_s32_x4(svcount_t pn, svint32x4_t zn1, svint32x4_t zn2, svint32x4_t zn3, svint32x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s32_x4)(pn, zn1, zn2);
@@ -345,45 +275,35 @@ svint32x4_t test_svsel_s32_x4(svcount_t pn, svint32x4_t zn1, svint32x4_t zn2, sv
 
 // CHECK-LABEL: @test_svsel_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN1_COERCE0:%.*]], <vscale x 4 x i32> [[ZN1_COERCE1:%.*]], <vscale x 4 x i32> [[ZN1_COERCE2:%.*]], <vscale x 4 x i32> [[ZN1_COERCE3:%.*]], <vscale x 4 x i32> [[ZN2_COERCE0:%.*]], <vscale x 4 x i32> [[ZN2_COERCE1:%.*]], <vscale x 4 x i32> [[ZN2_COERCE2:%.*]], <vscale x 4 x i32> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_u32_x4u11__SVCount_t12svuint32x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN2]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sel.x4.nxv4i32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x i32> [[ZN1_COERCE0:%.*]], <vscale x 4 x i32> [[ZN1_COERCE1:%.*]], <vscale x 4 x i32> [[ZN1_COERCE2:%.*]], <vscale x 4 x i32> [[ZN1_COERCE3:%.*]], <vscale x 4 x i32> [[ZN2_COERCE0:%.*]], <vscale x 4 x i32> [[ZN2_COERCE1:%.*]], <vscale x 4 x i32> [[ZN2_COERCE2:%.*]], <vscale x 4 x i32> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svsel_u32_x4(svcount_t pn, svuint32x4_t zn1, svuint32x4_t zn2, svuint32x4_t zn3, svuint32x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u32_x4)(pn, zn1, zn2);
@@ -391,45 +311,35 @@ svuint32x4_t test_svsel_u32_x4(svcount_t pn, svuint32x4_t zn1, svuint32x4_t zn2,
 
 // CHECK-LABEL: @test_svsel_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x4.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x4.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[ZN1_COERCE0:%.*]], <vscale x 4 x float> [[ZN1_COERCE1:%.*]], <vscale x 4 x float> [[ZN1_COERCE2:%.*]], <vscale x 4 x float> [[ZN1_COERCE3:%.*]], <vscale x 4 x float> [[ZN2_COERCE0:%.*]], <vscale x 4 x float> [[ZN2_COERCE1:%.*]], <vscale x 4 x float> [[ZN2_COERCE2:%.*]], <vscale x 4 x float> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_f32_x4u11__SVCount_t13svfloat32x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN2]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x4.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.sel.x4.nxv4f32(target("aarch64.svcount") [[PN:%.*]], <vscale x 4 x float> [[ZN1_COERCE0:%.*]], <vscale x 4 x float> [[ZN1_COERCE1:%.*]], <vscale x 4 x float> [[ZN1_COERCE2:%.*]], <vscale x 4 x float> [[ZN1_COERCE3:%.*]], <vscale x 4 x float> [[ZN2_COERCE0:%.*]], <vscale x 4 x float> [[ZN2_COERCE1:%.*]], <vscale x 4 x float> [[ZN2_COERCE2:%.*]], <vscale x 4 x float> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svsel_f32_x4(svcount_t pn, svfloat32x4_t zn1, svfloat32x4_t zn2, svfloat32x4_t zn3, svfloat32x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_f32_x4)(pn, zn1, zn2);
@@ -439,45 +349,35 @@ svfloat32x4_t test_svsel_f32_x4(svcount_t pn, svfloat32x4_t zn1, svfloat32x4_t z
 
 // CHECK-LABEL: @test_svsel_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN1_COERCE0:%.*]], <vscale x 2 x i64> [[ZN1_COERCE1:%.*]], <vscale x 2 x i64> [[ZN1_COERCE2:%.*]], <vscale x 2 x i64> [[ZN1_COERCE3:%.*]], <vscale x 2 x i64> [[ZN2_COERCE0:%.*]], <vscale x 2 x i64> [[ZN2_COERCE1:%.*]], <vscale x 2 x i64> [[ZN2_COERCE2:%.*]], <vscale x 2 x i64> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_s64_x4u11__SVCount_t11svint64x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN1_COERCE0:%.*]], <vscale x 2 x i64> [[ZN1_COERCE1:%.*]], <vscale x 2 x i64> [[ZN1_COERCE2:%.*]], <vscale x 2 x i64> [[ZN1_COERCE3:%.*]], <vscale x 2 x i64> [[ZN2_COERCE0:%.*]], <vscale x 2 x i64> [[ZN2_COERCE1:%.*]], <vscale x 2 x i64> [[ZN2_COERCE2:%.*]], <vscale x 2 x i64> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svsel_s64_x4(svcount_t pn, svint64x4_t zn1, svint64x4_t zn2, svint64x4_t zn3, svint64x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_s64_x4)(pn, zn1, zn2);
@@ -485,45 +385,35 @@ svint64x4_t test_svsel_s64_x4(svcount_t pn, svint64x4_t zn1, svint64x4_t zn2, sv
 
 // CHECK-LABEL: @test_svsel_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN1_COERCE0:%.*]], <vscale x 2 x i64> [[ZN1_COERCE1:%.*]], <vscale x 2 x i64> [[ZN1_COERCE2:%.*]], <vscale x 2 x i64> [[ZN1_COERCE3:%.*]], <vscale x 2 x i64> [[ZN2_COERCE0:%.*]], <vscale x 2 x i64> [[ZN2_COERCE1:%.*]], <vscale x 2 x i64> [[ZN2_COERCE2:%.*]], <vscale x 2 x i64> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_u64_x4u11__SVCount_t12svuint64x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN2]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sel.x4.nxv2i64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x i64> [[ZN1_COERCE0:%.*]], <vscale x 2 x i64> [[ZN1_COERCE1:%.*]], <vscale x 2 x i64> [[ZN1_COERCE2:%.*]], <vscale x 2 x i64> [[ZN1_COERCE3:%.*]], <vscale x 2 x i64> [[ZN2_COERCE0:%.*]], <vscale x 2 x i64> [[ZN2_COERCE1:%.*]], <vscale x 2 x i64> [[ZN2_COERCE2:%.*]], <vscale x 2 x i64> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svsel_u64_x4(svcount_t pn, svuint64x4_t zn1, svuint64x4_t zn2, svuint64x4_t zn3, svuint64x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_u64_x4)(pn, zn1, zn2);
@@ -531,45 +421,35 @@ svuint64x4_t test_svsel_u64_x4(svcount_t pn, svuint64x4_t zn1, svuint64x4_t zn2,
 
 // CHECK-LABEL: @test_svsel_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x4.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x4.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[ZN1_COERCE0:%.*]], <vscale x 2 x double> [[ZN1_COERCE1:%.*]], <vscale x 2 x double> [[ZN1_COERCE2:%.*]], <vscale x 2 x double> [[ZN1_COERCE3:%.*]], <vscale x 2 x double> [[ZN2_COERCE0:%.*]], <vscale x 2 x double> [[ZN2_COERCE1:%.*]], <vscale x 2 x double> [[ZN2_COERCE2:%.*]], <vscale x 2 x double> [[ZN2_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svsel_f64_x4u11__SVCount_t13svfloat64x4_tS0_S0_S0_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN2]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x4.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.sel.x4.nxv2f64(target("aarch64.svcount") [[PN:%.*]], <vscale x 2 x double> [[ZN1_COERCE0:%.*]], <vscale x 2 x double> [[ZN1_COERCE1:%.*]], <vscale x 2 x double> [[ZN1_COERCE2:%.*]], <vscale x 2 x double> [[ZN1_COERCE3:%.*]], <vscale x 2 x double> [[ZN2_COERCE0:%.*]], <vscale x 2 x double> [[ZN2_COERCE1:%.*]], <vscale x 2 x double> [[ZN2_COERCE2:%.*]], <vscale x 2 x double> [[ZN2_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svsel_f64_x4(svcount_t pn, svfloat64x4_t zn1, svfloat64x4_t zn2, svfloat64x4_t zn3, svfloat64x4_t zn4) __arm_streaming {
   return SVE_ACLE_FUNC(svsel,_f64_x4)(pn, zn1, zn2);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx2.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx2.c
index 4b523fb544d252..de605bab67cc35 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx2.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx2.c
@@ -20,25 +20,27 @@
 
 // CHECK-LABEL: @test_svuzp_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svuzp_s8_x210svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svuzp_s8_x2(svint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s8_x2)(zn);
@@ -46,25 +48,27 @@ svint8x2_t test_svuzp_s8_x2(svint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svuzp_u8_x211svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svuzp_u8_x2(svuint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u8_x2)(zn);
@@ -74,25 +78,27 @@ svuint8x2_t test_svuzp_u8_x2(svuint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_s16_x211svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svuzp_s16_x2(svint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s16_x2)(zn);
@@ -100,25 +106,27 @@ svint16x2_t test_svuzp_s16_x2(svint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_u16_x212svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svuzp_u16_x2(svuint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u16_x2)(zn);
@@ -126,25 +134,27 @@ svuint16x2_t test_svuzp_u16_x2(svuint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_f16_x213svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svuzp_f16_x2(svfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_f16_x2)(zn);
@@ -152,25 +162,27 @@ svfloat16x2_t test_svuzp_f16_x2(svfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzp_bf16_x214svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svuzp_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_bf16_x2)(zn);
@@ -180,25 +192,27 @@ svbfloat16x2_t test_svuzp_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svuzp_s32_x2(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s32_x2)(zn);
@@ -206,25 +220,27 @@ svint32x2_t test_svuzp_s32_x2(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_u32_x212svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svuzp_u32_x2(svuint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u32_x2)(zn);
@@ -232,25 +248,27 @@ svuint32x2_t test_svuzp_u32_x2(svuint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svuzp_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_f32_x2)(zn);
@@ -260,25 +278,27 @@ svfloat32x2_t test_svuzp_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_s64_x211svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svuzp_s64_x2(svint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s64_x2)(zn);
@@ -286,25 +306,27 @@ svint64x2_t test_svuzp_s64_x2(svint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_u64_x212svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svuzp_u64_x2(svuint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u64_x2)(zn);
@@ -312,25 +334,27 @@ svuint64x2_t test_svuzp_u64_x2(svuint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_f64_x213svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svuzp_f64_x2(svfloat64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_f64_x2)(zn);
@@ -340,25 +364,27 @@ svfloat64x2_t test_svuzp_f64_x2(svfloat64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzpq_s8_x210svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svuzpq_s8_x2(svint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s8_x2)(zn);
@@ -366,25 +392,27 @@ svint8x2_t test_svuzpq_s8_x2(svint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzpq_u8_x211svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svuzpq_u8_x2(svuint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u8_x2)(zn);
@@ -392,25 +420,27 @@ svuint8x2_t test_svuzpq_u8_x2(svuint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_s16_x211svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svuzpq_s16_x2(svint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s16_x2)(zn);
@@ -418,25 +448,27 @@ svint16x2_t test_svuzpq_s16_x2(svint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_u16_x212svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svuzpq_u16_x2(svuint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u16_x2)(zn);
@@ -444,25 +476,27 @@ svuint16x2_t test_svuzpq_u16_x2(svuint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_f16_x213svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svuzpq_f16_x2(svfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_f16_x2)(zn);
@@ -470,25 +504,27 @@ svfloat16x2_t test_svuzpq_f16_x2(svfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svuzpq_bf16_x214svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svuzpq_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_bf16_x2)(zn);
@@ -496,25 +532,27 @@ svbfloat16x2_t test_svuzpq_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svuzpq_s32_x2(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s32_x2)(zn);
@@ -522,25 +560,27 @@ svint32x2_t test_svuzpq_s32_x2(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_u32_x212svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svuzpq_u32_x2(svuint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u32_x2)(zn);
@@ -548,25 +588,27 @@ svuint32x2_t test_svuzpq_u32_x2(svuint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svuzpq_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_f32_x2)(zn);
@@ -574,25 +616,27 @@ svfloat32x2_t test_svuzpq_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_s64_x211svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svuzpq_s64_x2(svint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s64_x2)(zn);
@@ -600,25 +644,27 @@ svint64x2_t test_svuzpq_s64_x2(svint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_u64_x212svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svuzpq_u64_x2(svuint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u64_x2)(zn);
@@ -626,25 +672,27 @@ svuint64x2_t test_svuzpq_u64_x2(svuint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_f64_x213svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svuzpq_f64_x2(svfloat64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_f64_x2)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx4.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx4.c
index 8f8b7fbb5bbf46..aa210f59508b59 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx4.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_uzpx4.c
@@ -20,37 +20,35 @@
 
 // CHECK-LABEL: @test_svuzp_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svuzp_s8_x410svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svuzp_s8_x4(svint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s8_x4)(zn);
@@ -58,37 +56,35 @@ svint8x4_t test_svuzp_s8_x4(svint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svuzp_u8_x411svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzp.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svuzp_u8_x4(svuint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u8_x4)(zn);
@@ -98,37 +94,35 @@ svuint8x4_t test_svuzp_u8_x4(svuint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_s16_x411svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svuzp_s16_x4(svint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s16_x4)(zn);
@@ -136,37 +130,35 @@ svint16x4_t test_svuzp_s16_x4(svint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_u16_x412svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzp.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svuzp_u16_x4(svuint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u16_x4)(zn);
@@ -174,37 +166,35 @@ svuint16x4_t test_svuzp_u16_x4(svuint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_f16_x413svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzp.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svuzp_f16_x4(svfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_f16_x4)(zn);
@@ -212,37 +202,35 @@ svfloat16x4_t test_svuzp_f16_x4(svfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzp_bf16_x414svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzp.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svuzp_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_bf16_x4)(zn);
@@ -252,37 +240,35 @@ svbfloat16x4_t test_svuzp_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svuzp_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s32_x4)(zn);
@@ -290,37 +276,35 @@ svint32x4_t test_svuzp_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzp.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svuzp_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u32_x4)(zn);
@@ -328,37 +312,35 @@ svuint32x4_t test_svuzp_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzp.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svuzp_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_f32_x4)(zn);
@@ -368,37 +350,35 @@ svfloat32x4_t test_svuzp_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svuzp_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_s64_x4)(zn);
@@ -406,37 +386,35 @@ svint64x4_t test_svuzp_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzp.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svuzp_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_u64_x4)(zn);
@@ -444,37 +422,35 @@ svuint64x4_t test_svuzp_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzp_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzp_f64_x413svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzp.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svuzp_f64_x4(svfloat64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzp,_f64_x4)(zn);
@@ -484,37 +460,35 @@ svfloat64x4_t test_svuzp_f64_x4(svfloat64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzpq_s8_x410svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svuzpq_s8_x4(svint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s8_x4)(zn);
@@ -522,37 +496,35 @@ svint8x4_t test_svuzpq_s8_x4(svint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svuzpq_u8_x411svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.uzpq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svuzpq_u8_x4(svuint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u8_x4)(zn);
@@ -560,37 +532,35 @@ svuint8x4_t test_svuzpq_u8_x4(svuint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_s16_x411svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svuzpq_s16_x4(svint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s16_x4)(zn);
@@ -598,37 +568,35 @@ svint16x4_t test_svuzpq_s16_x4(svint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_u16_x412svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uzpq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svuzpq_u16_x4(svuint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u16_x4)(zn);
@@ -636,37 +604,35 @@ svuint16x4_t test_svuzpq_u16_x4(svuint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_f16_x413svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.uzpq.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svuzpq_f16_x4(svfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_f16_x4)(zn);
@@ -674,37 +640,35 @@ svfloat16x4_t test_svuzpq_f16_x4(svfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svuzpq_bf16_x414svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.uzpq.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svuzpq_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_bf16_x4)(zn);
@@ -712,37 +676,35 @@ svbfloat16x4_t test_svuzpq_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svuzpq_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s32_x4)(zn);
@@ -750,37 +712,35 @@ svint32x4_t test_svuzpq_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uzpq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svuzpq_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u32_x4)(zn);
@@ -788,37 +748,35 @@ svuint32x4_t test_svuzpq_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.uzpq.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svuzpq_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_f32_x4)(zn);
@@ -826,37 +784,35 @@ svfloat32x4_t test_svuzpq_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svuzpq_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_s64_x4)(zn);
@@ -864,37 +820,35 @@ svint64x4_t test_svuzpq_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uzpq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svuzpq_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_u64_x4)(zn);
@@ -902,37 +856,35 @@ svuint64x4_t test_svuzpq_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svuzpq_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svuzpq_f64_x413svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.uzpq.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svuzpq_f64_x4(svfloat64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svuzpq,_f64_x4)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx2.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx2.c
index bc72cdb65e38b9..a29c347e3197f3 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx2.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx2.c
@@ -19,25 +19,27 @@
 
 // CHECK-LABEL: @test_svzip_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svzip_s8_x210svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svzip_s8_x2(svint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s8_x2)(zn);
@@ -45,25 +47,27 @@ svint8x2_t test_svzip_s8_x2(svint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svzip_u8_x211svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svzip_u8_x2(svuint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u8_x2)(zn);
@@ -73,25 +77,27 @@ svuint8x2_t test_svzip_u8_x2(svuint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_s16_x211svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svzip_s16_x2(svint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s16_x2)(zn);
@@ -99,25 +105,27 @@ svint16x2_t test_svzip_s16_x2(svint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_u16_x212svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svzip_u16_x2(svuint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u16_x2)(zn);
@@ -125,25 +133,27 @@ svuint16x2_t test_svzip_u16_x2(svuint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_f16_x213svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svzip_f16_x2(svfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_f16_x2)(zn);
@@ -151,25 +161,27 @@ svfloat16x2_t test_svzip_f16_x2(svfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzip_bf16_x214svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svzip_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_bf16_x2)(zn);
@@ -179,25 +191,27 @@ svbfloat16x2_t test_svzip_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svzip_s32_x2(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s32_x2)(zn);
@@ -205,25 +219,27 @@ svint32x2_t test_svzip_s32_x2(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_u32_x212svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svzip_u32_x2(svuint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u32_x2)(zn);
@@ -231,25 +247,27 @@ svuint32x2_t test_svzip_u32_x2(svuint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svzip_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_f32_x2)(zn);
@@ -259,25 +277,27 @@ svfloat32x2_t test_svzip_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_s64_x211svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svzip_s64_x2(svint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s64_x2)(zn);
@@ -285,25 +305,27 @@ svint64x2_t test_svzip_s64_x2(svint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_u64_x212svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svzip_u64_x2(svuint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u64_x2)(zn);
@@ -311,25 +333,27 @@ svuint64x2_t test_svzip_u64_x2(svuint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_f64_x213svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svzip_f64_x2(svfloat64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_f64_x2)(zn);
@@ -339,25 +363,27 @@ svfloat64x2_t test_svzip_f64_x2(svfloat64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzipq_s8_x210svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svzipq_s8_x2(svint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s8_x2)(zn);
@@ -365,25 +391,27 @@ svint8x2_t test_svzipq_s8_x2(svint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u8_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzipq_u8_x211svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x2.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svzipq_u8_x2(svuint8x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u8_x2)(zn);
@@ -391,25 +419,27 @@ svuint8x2_t test_svzipq_u8_x2(svuint8x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_s16_x211svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svzipq_s16_x2(svint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s16_x2)(zn);
@@ -417,25 +447,27 @@ svint16x2_t test_svzipq_s16_x2(svint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_u16_x212svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x2.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svzipq_u16_x2(svuint16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u16_x2)(zn);
@@ -443,25 +475,27 @@ svuint16x2_t test_svzipq_u16_x2(svuint16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_f16_x213svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x2.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svzipq_f16_x2(svfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_f16_x2)(zn);
@@ -469,25 +503,27 @@ svfloat16x2_t test_svzipq_f16_x2(svfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_bf16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svzipq_bf16_x214svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x2.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svzipq_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_bf16_x2)(zn);
@@ -495,25 +531,27 @@ svbfloat16x2_t test_svzipq_bf16_x2(svbfloat16x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_s32_x211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svzipq_s32_x2(svint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s32_x2)(zn);
@@ -521,25 +559,27 @@ svint32x2_t test_svzipq_s32_x2(svint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_u32_x212svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x2.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svzipq_u32_x2(svuint32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u32_x2)(zn);
@@ -547,25 +587,27 @@ svuint32x2_t test_svzipq_u32_x2(svuint32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_f32_x213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x2.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svzipq_f32_x2(svfloat32x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_f32_x2)(zn);
@@ -573,25 +615,27 @@ svfloat32x2_t test_svzipq_f32_x2(svfloat32x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_s64_x211svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svzipq_s64_x2(svint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s64_x2)(zn);
@@ -599,25 +643,27 @@ svint64x2_t test_svzipq_s64_x2(svint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_u64_x212svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x2.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svzipq_u64_x2(svuint64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u64_x2)(zn);
@@ -625,25 +671,27 @@ svuint64x2_t test_svzipq_u64_x2(svuint64x2_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_f64_x213svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x2.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svzipq_f64_x2(svfloat64x2_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_f64_x2)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx4.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx4.c
index e4ad0bbda97669..be40ecb4bcaa35 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx4.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_vector_zipx4.c
@@ -19,37 +19,35 @@
 
 // CHECK-LABEL: @test_svzip_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svzip_s8_x410svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svzip_s8_x4(svint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s8_x4)(zn);
@@ -57,37 +55,35 @@ svint8x4_t test_svzip_s8_x4(svint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z16test_svzip_u8_x411svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zip.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svzip_u8_x4(svuint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u8_x4)(zn);
@@ -97,37 +93,35 @@ svuint8x4_t test_svzip_u8_x4(svuint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_s16_x411svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svzip_s16_x4(svint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s16_x4)(zn);
@@ -135,37 +129,35 @@ svint16x4_t test_svzip_s16_x4(svint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_u16_x412svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zip.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svzip_u16_x4(svuint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u16_x4)(zn);
@@ -173,37 +165,35 @@ svuint16x4_t test_svzip_u16_x4(svuint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_f16_x413svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zip.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svzip_f16_x4(svfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_f16_x4)(zn);
@@ -211,37 +201,35 @@ svfloat16x4_t test_svzip_f16_x4(svfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzip_bf16_x414svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zip.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svzip_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_bf16_x4)(zn);
@@ -251,37 +239,35 @@ svbfloat16x4_t test_svzip_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svzip_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s32_x4)(zn);
@@ -289,37 +275,35 @@ svint32x4_t test_svzip_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zip.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svzip_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u32_x4)(zn);
@@ -327,37 +311,35 @@ svuint32x4_t test_svzip_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zip.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svzip_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_f32_x4)(zn);
@@ -367,37 +349,35 @@ svfloat32x4_t test_svzip_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svzip_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_s64_x4)(zn);
@@ -405,37 +385,35 @@ svint64x4_t test_svzip_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zip.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svzip_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_u64_x4)(zn);
@@ -443,37 +421,35 @@ svuint64x4_t test_svzip_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzip_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzip_f64_x413svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zip.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svzip_f64_x4(svfloat64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzip,_f64_x4)(zn);
@@ -483,37 +459,35 @@ svfloat64x4_t test_svzip_f64_x4(svfloat64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzipq_s8_x410svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svzipq_s8_x4(svint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s8_x4)(zn);
@@ -521,37 +495,35 @@ svint8x4_t test_svzipq_s8_x4(svint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u8_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svzipq_u8_x411svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[ZN]], i64 48)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP8]], <vscale x 16 x i8> [[TMP9]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP10]], <vscale x 16 x i8> [[TMP11]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.zipq.x4.nxv16i8(<vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svzipq_u8_x4(svuint8x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u8_x4)(zn);
@@ -559,37 +531,35 @@ svuint8x4_t test_svzipq_u8_x4(svuint8x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_s16_x411svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svzipq_s16_x4(svint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s16_x4)(zn);
@@ -597,37 +567,35 @@ svint16x4_t test_svzipq_s16_x4(svint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_u16_x412svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP8]], <vscale x 8 x i16> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP10]], <vscale x 8 x i16> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.zipq.x4.nxv8i16(<vscale x 8 x i16> [[ZN_COERCE0:%.*]], <vscale x 8 x i16> [[ZN_COERCE1:%.*]], <vscale x 8 x i16> [[ZN_COERCE2:%.*]], <vscale x 8 x i16> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svzipq_u16_x4(svuint16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u16_x4)(zn);
@@ -635,37 +603,35 @@ svuint16x4_t test_svzipq_u16_x4(svuint16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_f16_x413svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.zipq.x4.nxv8f16(<vscale x 8 x half> [[ZN_COERCE0:%.*]], <vscale x 8 x half> [[ZN_COERCE1:%.*]], <vscale x 8 x half> [[ZN_COERCE2:%.*]], <vscale x 8 x half> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svzipq_f16_x4(svfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_f16_x4)(zn);
@@ -673,37 +639,35 @@ svfloat16x4_t test_svzipq_f16_x4(svfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_bf16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svzipq_bf16_x414svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.zipq.x4.nxv8bf16(<vscale x 8 x bfloat> [[ZN_COERCE0:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE1:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE2:%.*]], <vscale x 8 x bfloat> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svzipq_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_bf16_x4)(zn);
@@ -711,37 +675,35 @@ svbfloat16x4_t test_svzipq_bf16_x4(svbfloat16x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_s32_x411svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svzipq_s32_x4(svint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s32_x4)(zn);
@@ -749,37 +711,35 @@ svint32x4_t test_svzipq_s32_x4(svint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_u32_x412svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.zipq.x4.nxv4i32(<vscale x 4 x i32> [[ZN_COERCE0:%.*]], <vscale x 4 x i32> [[ZN_COERCE1:%.*]], <vscale x 4 x i32> [[ZN_COERCE2:%.*]], <vscale x 4 x i32> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svzipq_u32_x4(svuint32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u32_x4)(zn);
@@ -787,37 +747,35 @@ svuint32x4_t test_svzipq_u32_x4(svuint32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_f32_x413svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZN]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.zipq.x4.nxv4f32(<vscale x 4 x float> [[ZN_COERCE0:%.*]], <vscale x 4 x float> [[ZN_COERCE1:%.*]], <vscale x 4 x float> [[ZN_COERCE2:%.*]], <vscale x 4 x float> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svzipq_f32_x4(svfloat32x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_f32_x4)(zn);
@@ -825,37 +783,35 @@ svfloat32x4_t test_svzipq_f32_x4(svfloat32x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_s64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_s64_x411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svzipq_s64_x4(svint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_s64_x4)(zn);
@@ -863,37 +819,35 @@ svint64x4_t test_svzipq_s64_x4(svint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_u64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_u64_x412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP8]], <vscale x 2 x i64> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.zipq.x4.nxv2i64(<vscale x 2 x i64> [[ZN_COERCE0:%.*]], <vscale x 2 x i64> [[ZN_COERCE1:%.*]], <vscale x 2 x i64> [[ZN_COERCE2:%.*]], <vscale x 2 x i64> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svzipq_u64_x4(svuint64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_u64_x4)(zn);
@@ -901,37 +855,35 @@ svuint64x4_t test_svzipq_u64_x4(svuint64x4_t zn) __arm_streaming {
 
 // CHECK-LABEL: @test_svzipq_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svzipq_f64_x413svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZN]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.zipq.x4.nxv2f64(<vscale x 2 x double> [[ZN_COERCE0:%.*]], <vscale x 2 x double> [[ZN_COERCE1:%.*]], <vscale x 2 x double> [[ZN_COERCE2:%.*]], <vscale x 2 x double> [[ZN_COERCE3:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 2
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svzipq_f64_x4(svfloat64x4_t zn) __arm_streaming {
   return SVE_ACLE_FUNC(svzipq,_f64_x4)(zn);

diff  --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_write.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_write.c
index e52301aab455bb..1bdca4a12bcbd9 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_write.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_write.c
@@ -18,16 +18,12 @@
 
 // CHECK-LABEL: @test_svwrite_ver_za8_u8_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_ver_za8_u8_vg2j11svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za8_u8_vg2(uint32_t base, svuint8x2_t val) __arm_streaming __arm_inout("za") {
@@ -36,16 +32,12 @@ void test_svwrite_ver_za8_u8_vg2(uint32_t base, svuint8x2_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_ver_za8_s8_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_ver_za8_s8_vg2j10svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za8_s8_vg2(uint32_t base, svint8x2_t val) __arm_streaming __arm_inout("za") {
@@ -54,16 +46,12 @@ void test_svwrite_ver_za8_s8_vg2(uint32_t base, svint8x2_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_hor_za8_u8_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_hor_za8_u8_vg2j11svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za8_u8_vg2(uint32_t base, svuint8x2_t val) __arm_streaming __arm_inout("za") {
@@ -72,16 +60,12 @@ void test_svwrite_hor_za8_u8_vg2(uint32_t base, svuint8x2_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_hor_za8_s8_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_hor_za8_s8_vg2j10svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za8_s8_vg2(uint32_t base, svint8x2_t val) __arm_streaming __arm_inout("za") {
@@ -90,20 +74,12 @@ void test_svwrite_hor_za8_s8_vg2(uint32_t base, svint8x2_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_hor_za8_u8_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_hor_za8_u8_vg4j11svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za8_u8_vg4(uint32_t base, svuint8x4_t val) __arm_streaming __arm_inout("za") {
@@ -112,20 +88,12 @@ void test_svwrite_hor_za8_u8_vg4(uint32_t base, svuint8x4_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_hor_za8_s8_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_hor_za8_s8_vg4j10svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za8_s8_vg4(uint32_t base, svint8x4_t val) __arm_streaming __arm_inout("za") {
@@ -134,20 +102,12 @@ void test_svwrite_hor_za8_s8_vg4(uint32_t base, svint8x4_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_ver_za8_u8_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_ver_za8_u8_vg4j11svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za8_u8_vg4(uint32_t base, svuint8x4_t val) __arm_streaming __arm_inout("za") {
@@ -156,20 +116,12 @@ void test_svwrite_ver_za8_u8_vg4(uint32_t base, svuint8x4_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_ver_za8_s8_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_ver_za8_s8_vg4j10svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv16i8(i32 0, i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za8_s8_vg4(uint32_t base, svint8x4_t val) __arm_streaming __arm_inout("za") {
@@ -178,16 +130,12 @@ void test_svwrite_ver_za8_s8_vg4(uint32_t base, svint8x4_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_hor_za16_u16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za16_u16_vg2j12svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_u16_vg2(uint32_t base, svuint16x2_t val) __arm_streaming __arm_inout("za") {
@@ -196,16 +144,12 @@ void test_svwrite_hor_za16_u16_vg2(uint32_t base, svuint16x2_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_hor_za16_bf16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svwrite_hor_za16_bf16_vg2j14svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_bf16_vg2(uint32_t base, svbfloat16x2_t val) __arm_streaming __arm_inout("za") {
@@ -214,16 +158,12 @@ void test_svwrite_hor_za16_bf16_vg2(uint32_t base, svbfloat16x2_t val) __arm_str
 
 // CHECK-LABEL: @test_svwrite_hor_za16_f16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za16_f16_vg2j13svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_f16_vg2(uint32_t base, svfloat16x2_t val) __arm_streaming __arm_inout("za") {
@@ -232,16 +172,12 @@ void test_svwrite_hor_za16_f16_vg2(uint32_t base, svfloat16x2_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_hor_za16_s16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za16_s16_vg2j11svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_s16_vg2(uint32_t base, svint16x2_t val) __arm_streaming __arm_inout("za") {
@@ -250,16 +186,12 @@ void test_svwrite_hor_za16_s16_vg2(uint32_t base, svint16x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_ver_za16_u16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za16_u16_vg2j12svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_u16_vg2(uint32_t base, svuint16x2_t val) __arm_streaming __arm_inout("za") {
@@ -268,16 +200,12 @@ void test_svwrite_ver_za16_u16_vg2(uint32_t base, svuint16x2_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_ver_za16_bf16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svwrite_ver_za16_bf16_vg2j14svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_bf16_vg2(uint32_t base, svbfloat16x2_t val) __arm_streaming __arm_inout("za") {
@@ -286,16 +214,12 @@ void test_svwrite_ver_za16_bf16_vg2(uint32_t base, svbfloat16x2_t val) __arm_str
 
 // CHECK-LABEL: @test_svwrite_ver_za16_f16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za16_f16_vg2j13svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_f16_vg2(uint32_t base, svfloat16x2_t val) __arm_streaming __arm_inout("za") {
@@ -304,16 +228,12 @@ void test_svwrite_ver_za16_f16_vg2(uint32_t base, svfloat16x2_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_ver_za16_s16_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za16_s16_vg2j11svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_s16_vg2(uint32_t base, svint16x2_t val) __arm_streaming __arm_inout("za") {
@@ -322,20 +242,12 @@ void test_svwrite_ver_za16_s16_vg2(uint32_t base, svint16x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_hor_za16_u16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za16_u16_vg4j12svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_u16_vg4(uint32_t base, svuint16x4_t val) __arm_streaming __arm_inout("za") {
@@ -344,20 +256,12 @@ void test_svwrite_hor_za16_u16_vg4(uint32_t base, svuint16x4_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_hor_za16_bf16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE2:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svwrite_hor_za16_bf16_vg4j14svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE2:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_bf16_vg4(uint32_t base, svbfloat16x4_t val) __arm_streaming __arm_inout("za") {
@@ -366,20 +270,12 @@ void test_svwrite_hor_za16_bf16_vg4(uint32_t base, svbfloat16x4_t val) __arm_str
 
 // CHECK-LABEL: @test_svwrite_hor_za16_f16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]], <vscale x 8 x half> [[VAL_COERCE2:%.*]], <vscale x 8 x half> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za16_f16_vg4j13svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]], <vscale x 8 x half> [[VAL_COERCE2:%.*]], <vscale x 8 x half> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_f16_vg4(uint32_t base, svfloat16x4_t val) __arm_streaming __arm_inout("za") {
@@ -388,20 +284,12 @@ void test_svwrite_hor_za16_f16_vg4(uint32_t base, svfloat16x4_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_hor_za16_s16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za16_s16_vg4j11svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za16_s16_vg4(uint32_t base, svint16x4_t val) __arm_streaming __arm_inout("za") {
@@ -410,20 +298,12 @@ void test_svwrite_hor_za16_s16_vg4(uint32_t base, svint16x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_ver_za16_u16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za16_u16_vg4j12svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_u16_vg4(uint32_t base, svuint16x4_t val) __arm_streaming __arm_inout("za") {
@@ -432,20 +312,12 @@ void test_svwrite_ver_za16_u16_vg4(uint32_t base, svuint16x4_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_ver_za16_bf16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE2:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z30test_svwrite_ver_za16_bf16_vg4j14svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8bf16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE2:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_bf16_vg4(uint32_t base, svbfloat16x4_t val) __arm_streaming __arm_inout("za") {
@@ -454,20 +326,12 @@ void test_svwrite_ver_za16_bf16_vg4(uint32_t base, svbfloat16x4_t val) __arm_str
 
 // CHECK-LABEL: @test_svwrite_ver_za16_f16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]], <vscale x 8 x half> [[VAL_COERCE2:%.*]], <vscale x 8 x half> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za16_f16_vg4j13svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8f16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]], <vscale x 8 x half> [[VAL_COERCE2:%.*]], <vscale x 8 x half> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_f16_vg4(uint32_t base, svfloat16x4_t val) __arm_streaming __arm_inout("za") {
@@ -476,20 +340,12 @@ void test_svwrite_ver_za16_f16_vg4(uint32_t base, svfloat16x4_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_ver_za16_s16_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za16_s16_vg4j11svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv8i16(i32 1, i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za16_s16_vg4(uint32_t base, svint16x4_t val) __arm_streaming __arm_inout("za") {
@@ -498,16 +354,12 @@ void test_svwrite_ver_za16_s16_vg4(uint32_t base, svint16x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_hor_za32_u32_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za32_u32_vg2j12svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za32_u32_vg2(uint32_t base, svuint32x2_t val) __arm_streaming __arm_inout("za") {
@@ -516,16 +368,12 @@ void test_svwrite_hor_za32_u32_vg2(uint32_t base, svuint32x2_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_hor_za32_f32_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za32_f32_vg2j13svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za32_f32_vg2(uint32_t base, svfloat32x2_t val) __arm_streaming __arm_inout("za") {
@@ -534,16 +382,12 @@ void test_svwrite_hor_za32_f32_vg2(uint32_t base, svfloat32x2_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_hor_za32_s32_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za32_s32_vg2j11svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za32_s32_vg2(uint32_t base, svint32x2_t val) __arm_streaming __arm_inout("za") {
@@ -552,16 +396,12 @@ void test_svwrite_hor_za32_s32_vg2(uint32_t base, svint32x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_ver_za32_u32_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za32_u32_vg2j12svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za32_u32_vg2(uint32_t base, svuint32x2_t val) __arm_streaming __arm_inout("za") {
@@ -570,16 +410,12 @@ void test_svwrite_ver_za32_u32_vg2(uint32_t base, svuint32x2_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_ver_za32_f32_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za32_f32_vg2j13svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za32_f32_vg2(uint32_t base, svfloat32x2_t val) __arm_streaming __arm_inout("za") {
@@ -588,16 +424,12 @@ void test_svwrite_ver_za32_f32_vg2(uint32_t base, svfloat32x2_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_ver_za32_s32_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za32_s32_vg2j11svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za32_s32_vg2(uint32_t base, svint32x2_t val) __arm_streaming __arm_inout("za") {
@@ -606,20 +438,12 @@ void test_svwrite_ver_za32_s32_vg2(uint32_t base, svint32x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_hor_za32_u32_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za32_u32_vg4j12svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za32_u32_vg4(uint32_t base, svuint32x4_t val) __arm_streaming __arm_inout("za") {
@@ -628,20 +452,12 @@ void test_svwrite_hor_za32_u32_vg4(uint32_t base, svuint32x4_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_hor_za32_f32_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]], <vscale x 4 x float> [[VAL_COERCE2:%.*]], <vscale x 4 x float> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za32_f32_vg4j13svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]], <vscale x 4 x float> [[VAL_COERCE2:%.*]], <vscale x 4 x float> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za32_f32_vg4(uint32_t base, svfloat32x4_t val) __arm_streaming __arm_inout("za") {
@@ -650,20 +466,12 @@ void test_svwrite_hor_za32_f32_vg4(uint32_t base, svfloat32x4_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_hor_za32_s32_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za32_s32_vg4j11svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za32_s32_vg4(uint32_t base, svint32x4_t val) __arm_streaming __arm_inout("za") {
@@ -672,20 +480,12 @@ void test_svwrite_hor_za32_s32_vg4(uint32_t base, svint32x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_ver_za32_u32_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za32_u32_vg4j12svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za32_u32_vg4(uint32_t base, svuint32x4_t val) __arm_streaming __arm_inout("za") {
@@ -694,20 +494,12 @@ void test_svwrite_ver_za32_u32_vg4(uint32_t base, svuint32x4_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_ver_za32_f32_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]], <vscale x 4 x float> [[VAL_COERCE2:%.*]], <vscale x 4 x float> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za32_f32_vg4j13svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4f32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]], <vscale x 4 x float> [[VAL_COERCE2:%.*]], <vscale x 4 x float> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za32_f32_vg4(uint32_t base, svfloat32x4_t val) __arm_streaming __arm_inout("za") {
@@ -716,20 +508,12 @@ void test_svwrite_ver_za32_f32_vg4(uint32_t base, svfloat32x4_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_ver_za32_s32_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za32_s32_vg4j11svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv4i32(i32 3, i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za32_s32_vg4(uint32_t base, svint32x4_t val) __arm_streaming __arm_inout("za") {
@@ -738,16 +522,12 @@ void test_svwrite_ver_za32_s32_vg4(uint32_t base, svint32x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_hor_za64_u64_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za64_u64_vg2j12svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za64_u64_vg2(uint32_t base, svuint64x2_t val) __arm_streaming __arm_inout("za") {
@@ -756,16 +536,12 @@ void test_svwrite_hor_za64_u64_vg2(uint32_t base, svuint64x2_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_hor_za64_f64_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za64_f64_vg2j13svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za64_f64_vg2(uint32_t base, svfloat64x2_t val) __arm_streaming __arm_inout("za") {
@@ -774,16 +550,12 @@ void test_svwrite_hor_za64_f64_vg2(uint32_t base, svfloat64x2_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_hor_za64_s64_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za64_s64_vg2j11svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za64_s64_vg2(uint32_t base, svint64x2_t val) __arm_streaming __arm_inout("za") {
@@ -792,16 +564,12 @@ void test_svwrite_hor_za64_s64_vg2(uint32_t base, svint64x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_ver_za64_u64_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za64_u64_vg2j12svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za64_u64_vg2(uint32_t base, svuint64x2_t val) __arm_streaming __arm_inout("za") {
@@ -810,16 +578,12 @@ void test_svwrite_ver_za64_u64_vg2(uint32_t base, svuint64x2_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_ver_za64_f64_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za64_f64_vg2j13svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za64_f64_vg2(uint32_t base, svfloat64x2_t val) __arm_streaming __arm_inout("za") {
@@ -828,16 +592,12 @@ void test_svwrite_ver_za64_f64_vg2(uint32_t base, svfloat64x2_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_ver_za64_s64_vg2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za64_s64_vg2j11svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg2.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za64_s64_vg2(uint32_t base, svint64x2_t val) __arm_streaming __arm_inout("za") {
@@ -846,20 +606,12 @@ void test_svwrite_ver_za64_s64_vg2(uint32_t base, svint64x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_hor_za64_u64_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za64_u64_vg4j12svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za64_u64_vg4(uint32_t base, svuint64x4_t val) __arm_streaming __arm_inout("za") {
@@ -868,20 +620,12 @@ void test_svwrite_hor_za64_u64_vg4(uint32_t base, svuint64x4_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_ver_za64_u64_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za64_u64_vg4j12svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za64_u64_vg4(uint32_t base, svuint64x4_t val) __arm_streaming __arm_inout("za") {
@@ -890,20 +634,12 @@ void test_svwrite_ver_za64_u64_vg4(uint32_t base, svuint64x4_t val) __arm_stream
 
 // CHECK-LABEL: @test_svwrite_hor_za64_f64_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]], <vscale x 2 x double> [[VAL_COERCE2:%.*]], <vscale x 2 x double> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za64_f64_vg4j13svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]], <vscale x 2 x double> [[VAL_COERCE2:%.*]], <vscale x 2 x double> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za64_f64_vg4(uint32_t base, svfloat64x4_t val) __arm_streaming __arm_inout("za") {
@@ -912,20 +648,12 @@ void test_svwrite_hor_za64_f64_vg4(uint32_t base, svfloat64x4_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_hor_za64_s64_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_hor_za64_s64_vg4j11svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.hor.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_hor_za64_s64_vg4(uint32_t base, svint64x4_t val) __arm_streaming __arm_inout("za") {
@@ -934,20 +662,12 @@ void test_svwrite_hor_za64_s64_vg4(uint32_t base, svint64x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_ver_za64_f64_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]], <vscale x 2 x double> [[VAL_COERCE2:%.*]], <vscale x 2 x double> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za64_f64_vg4j13svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2f64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]], <vscale x 2 x double> [[VAL_COERCE2:%.*]], <vscale x 2 x double> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za64_f64_vg4(uint32_t base, svfloat64x4_t val) __arm_streaming __arm_inout("za") {
@@ -956,20 +676,12 @@ void test_svwrite_ver_za64_f64_vg4(uint32_t base, svfloat64x4_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_ver_za64_s64_vg4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z29test_svwrite_ver_za64_s64_vg4j11svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.ver.vg4.nxv2i64(i32 7, i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_ver_za64_s64_vg4(uint32_t base, svint64x4_t val) __arm_streaming __arm_inout("za") {
@@ -978,16 +690,12 @@ void test_svwrite_ver_za64_s64_vg4(uint32_t base, svint64x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_za8_s8_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svwrite_za8_s8_vg1x2j10svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za8_s8_vg1x2(uint32_t base, svint8x2_t val) __arm_streaming __arm_inout("za") {
@@ -996,16 +704,12 @@ void test_svwrite_za8_s8_vg1x2(uint32_t base, svint8x2_t val) __arm_streaming __
 
 // CHECK-LABEL: @test_svwrite_za8_u8_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svwrite_za8_u8_vg1x2j11svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za8_u8_vg1x2(uint32_t base, svuint8x2_t val) __arm_streaming __arm_inout("za") {
@@ -1014,16 +718,12 @@ void test_svwrite_za8_u8_vg1x2(uint32_t base, svuint8x2_t val) __arm_streaming _
 
 // CHECK-LABEL: @test_svwrite_za16_s16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za16_s16_vg1x2j11svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_s16_vg1x2(uint32_t base, svint16x2_t val) __arm_streaming __arm_inout("za") {
@@ -1032,16 +732,12 @@ void test_svwrite_za16_s16_vg1x2(uint32_t base, svint16x2_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_za16_u16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za16_u16_vg1x2j12svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_u16_vg1x2(uint32_t base, svuint16x2_t val) __arm_streaming __arm_inout("za") {
@@ -1050,16 +746,12 @@ void test_svwrite_za16_u16_vg1x2(uint32_t base, svuint16x2_t val) __arm_streamin
 
 // CHECK-LABEL: @test_svwrite_za16_bf16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svwrite_za16_bf16_vg1x2j14svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_bf16_vg1x2(uint32_t base, svbfloat16x2_t val) __arm_streaming __arm_inout("za") {
@@ -1068,16 +760,12 @@ void test_svwrite_za16_bf16_vg1x2(uint32_t base, svbfloat16x2_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_za16_f16_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL]], i64 8)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za16_f16_vg1x2j13svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_f16_vg1x2(uint32_t base, svfloat16x2_t val) __arm_streaming __arm_inout("za") {
@@ -1086,16 +774,12 @@ void test_svwrite_za16_f16_vg1x2(uint32_t base, svfloat16x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_za32_s32_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za32_s32_vg1x2j11svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za32_s32_vg1x2(uint32_t base, svint32x2_t val) __arm_streaming __arm_inout("za") {
@@ -1104,16 +788,12 @@ void test_svwrite_za32_s32_vg1x2(uint32_t base, svint32x2_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_za32_u32_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za32_u32_vg1x2j12svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za32_u32_vg1x2(uint32_t base, svuint32x2_t val) __arm_streaming __arm_inout("za") {
@@ -1122,16 +802,12 @@ void test_svwrite_za32_u32_vg1x2(uint32_t base, svuint32x2_t val) __arm_streamin
 
 // CHECK-LABEL: @test_svwrite_za32_f32_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL]], i64 4)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za32_f32_vg1x2j13svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za32_f32_vg1x2(uint32_t base, svfloat32x2_t val) __arm_streaming __arm_inout("za") {
@@ -1140,16 +816,12 @@ void test_svwrite_za32_f32_vg1x2(uint32_t base, svfloat32x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_za64_u64_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za64_u64_vg1x2j12svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za64_u64_vg1x2(uint32_t base, svuint64x2_t val) __arm_streaming __arm_inout("za") {
@@ -1158,16 +830,12 @@ void test_svwrite_za64_u64_vg1x2(uint32_t base, svuint64x2_t val) __arm_streamin
 
 // CHECK-LABEL: @test_svwrite_za64_f64_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za64_f64_vg1x2j13svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za64_f64_vg1x2(uint32_t base, svfloat64x2_t val) __arm_streaming __arm_inout("za") {
@@ -1176,16 +844,12 @@ void test_svwrite_za64_f64_vg1x2(uint32_t base, svfloat64x2_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_za64_s64_vg1x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za64_s64_vg1x2j11svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x2.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za64_s64_vg1x2(uint32_t base, svint64x2_t val) __arm_streaming __arm_inout("za") {
@@ -1194,20 +858,12 @@ void test_svwrite_za64_s64_vg1x2(uint32_t base, svint64x2_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_za8_s8_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svwrite_za8_s8_vg1x4j10svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za8_s8_vg1x4(uint32_t base, svint8x4_t val) __arm_streaming __arm_inout("za") {
@@ -1216,20 +872,12 @@ void test_svwrite_za8_s8_vg1x4(uint32_t base, svint8x4_t val) __arm_streaming __
 
 // CHECK-LABEL: @test_svwrite_za8_u8_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z25test_svwrite_za8_u8_vg1x4j11svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[VAL]], i64 48)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[TMP0]], <vscale x 16 x i8> [[TMP1]], <vscale x 16 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv16i8(i32 [[BASE:%.*]], <vscale x 16 x i8> [[VAL_COERCE0:%.*]], <vscale x 16 x i8> [[VAL_COERCE1:%.*]], <vscale x 16 x i8> [[VAL_COERCE2:%.*]], <vscale x 16 x i8> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za8_u8_vg1x4(uint32_t base, svuint8x4_t val) __arm_streaming __arm_inout("za") {
@@ -1238,20 +886,12 @@ void test_svwrite_za8_u8_vg1x4(uint32_t base, svuint8x4_t val) __arm_streaming _
 
 // CHECK-LABEL: @test_svwrite_za16_s16_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za16_s16_vg1x4j11svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_s16_vg1x4(uint32_t base, svint16x4_t val) __arm_streaming __arm_inout("za") {
@@ -1260,20 +900,12 @@ void test_svwrite_za16_s16_vg1x4(uint32_t base, svint16x4_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_za16_u16_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za16_u16_vg1x4j12svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[TMP0]], <vscale x 8 x i16> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8i16(i32 [[BASE:%.*]], <vscale x 8 x i16> [[VAL_COERCE0:%.*]], <vscale x 8 x i16> [[VAL_COERCE1:%.*]], <vscale x 8 x i16> [[VAL_COERCE2:%.*]], <vscale x 8 x i16> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_u16_vg1x4(uint32_t base, svuint16x4_t val) __arm_streaming __arm_inout("za") {
@@ -1282,20 +914,12 @@ void test_svwrite_za16_u16_vg1x4(uint32_t base, svuint16x4_t val) __arm_streamin
 
 // CHECK-LABEL: @test_svwrite_za16_bf16_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE2:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z28test_svwrite_za16_bf16_vg1x4j14svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8bf16(i32 [[BASE:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE0:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE1:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE2:%.*]], <vscale x 8 x bfloat> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_bf16_vg1x4(uint32_t base, svbfloat16x4_t val) __arm_streaming __arm_inout("za") {
@@ -1304,20 +928,12 @@ void test_svwrite_za16_bf16_vg1x4(uint32_t base, svbfloat16x4_t val) __arm_strea
 
 // CHECK-LABEL: @test_svwrite_za16_f16_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 24)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]], <vscale x 8 x half> [[VAL_COERCE2:%.*]], <vscale x 8 x half> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za16_f16_vg1x4j13svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[VAL]], i64 24)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv8f16(i32 [[BASE:%.*]], <vscale x 8 x half> [[VAL_COERCE0:%.*]], <vscale x 8 x half> [[VAL_COERCE1:%.*]], <vscale x 8 x half> [[VAL_COERCE2:%.*]], <vscale x 8 x half> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za16_f16_vg1x4(uint32_t base, svfloat16x4_t val) __arm_streaming __arm_inout("za") {
@@ -1326,20 +942,12 @@ void test_svwrite_za16_f16_vg1x4(uint32_t base, svfloat16x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_za32_s32_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za32_s32_vg1x4j11svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za32_s32_vg1x4(uint32_t base, svint32x4_t val) __arm_streaming __arm_inout("za") {
@@ -1348,20 +956,12 @@ void test_svwrite_za32_s32_vg1x4(uint32_t base, svint32x4_t val) __arm_streaming
 
 // CHECK-LABEL: @test_svwrite_za32_u32_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za32_u32_vg1x4j12svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4i32(i32 [[BASE:%.*]], <vscale x 4 x i32> [[VAL_COERCE0:%.*]], <vscale x 4 x i32> [[VAL_COERCE1:%.*]], <vscale x 4 x i32> [[VAL_COERCE2:%.*]], <vscale x 4 x i32> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za32_u32_vg1x4(uint32_t base, svuint32x4_t val) __arm_streaming __arm_inout("za") {
@@ -1370,20 +970,12 @@ void test_svwrite_za32_u32_vg1x4(uint32_t base, svuint32x4_t val) __arm_streamin
 
 // CHECK-LABEL: @test_svwrite_za32_f32_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 12)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]], <vscale x 4 x float> [[VAL_COERCE2:%.*]], <vscale x 4 x float> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za32_f32_vg1x4j13svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[VAL]], i64 12)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv4f32(i32 [[BASE:%.*]], <vscale x 4 x float> [[VAL_COERCE0:%.*]], <vscale x 4 x float> [[VAL_COERCE1:%.*]], <vscale x 4 x float> [[VAL_COERCE2:%.*]], <vscale x 4 x float> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za32_f32_vg1x4(uint32_t base, svfloat32x4_t val) __arm_streaming __arm_inout("za") {
@@ -1392,20 +984,12 @@ void test_svwrite_za32_f32_vg1x4(uint32_t base, svfloat32x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_za64_u64_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za64_u64_vg1x4j12svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za64_u64_vg1x4(uint32_t base, svuint64x4_t val) __arm_streaming __arm_inout("za") {
@@ -1414,20 +998,12 @@ void test_svwrite_za64_u64_vg1x4(uint32_t base, svuint64x4_t val) __arm_streamin
 
 // CHECK-LABEL: @test_svwrite_za64_f64_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]], <vscale x 2 x double> [[VAL_COERCE2:%.*]], <vscale x 2 x double> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za64_f64_vg1x4j13svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2f64(i32 [[BASE:%.*]], <vscale x 2 x double> [[VAL_COERCE0:%.*]], <vscale x 2 x double> [[VAL_COERCE1:%.*]], <vscale x 2 x double> [[VAL_COERCE2:%.*]], <vscale x 2 x double> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za64_f64_vg1x4(uint32_t base, svfloat64x4_t val) __arm_streaming __arm_inout("za") {
@@ -1436,20 +1012,12 @@ void test_svwrite_za64_f64_vg1x4(uint32_t base, svfloat64x4_t val) __arm_streami
 
 // CHECK-LABEL: @test_svwrite_za64_s64_vg1x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CHECK-NEXT:    ret void
 //
 // CPP-CHECK-LABEL: @_Z27test_svwrite_za64_s64_vg1x4j11svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[VAL]], i64 6)
-// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[TMP0]], <vscale x 2 x i64> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
+// CPP-CHECK-NEXT:    tail call void @llvm.aarch64.sme.write.vg1x4.nxv2i64(i32 [[BASE:%.*]], <vscale x 2 x i64> [[VAL_COERCE0:%.*]], <vscale x 2 x i64> [[VAL_COERCE1:%.*]], <vscale x 2 x i64> [[VAL_COERCE2:%.*]], <vscale x 2 x i64> [[VAL_COERCE3:%.*]])
 // CPP-CHECK-NEXT:    ret void
 //
 void test_svwrite_za64_s64_vg1x4(uint32_t base, svint64x4_t val) __arm_streaming __arm_inout("za") {

diff  --git a/clang/test/CodeGen/aarch64-sme2p1-intrinsics/acle_sme2p1_movaz.c b/clang/test/CodeGen/aarch64-sme2p1-intrinsics/acle_sme2p1_movaz.c
index 466ca130326335..77b02b4c4708fa 100644
--- a/clang/test/CodeGen/aarch64-sme2p1-intrinsics/acle_sme2p1_movaz.c
+++ b/clang/test/CodeGen/aarch64-sme2p1-intrinsics/acle_sme2p1_movaz.c
@@ -8,150 +8,186 @@
 
 //
 // X2- hor
-// CHECK-LABEL: define dso_local <vscale x 32 x i8> @test_svreadz_hor_za8_s8_x2(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_hor_za8_s8_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0:[0-9]+]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i8> @_Z26test_svreadz_hor_za8_s8_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_hor_za8_s8_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0:[0-9]+]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svreadz_hor_za8_s8_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za8_s8_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i8> @test_svreadz_hor_za8_u8_x2(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_hor_za8_u8_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i8> @_Z26test_svreadz_hor_za8_u8_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_hor_za8_u8_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svreadz_hor_za8_u8_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za8_u8_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i16> @test_svreadz_hor_za16_s16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_hor_za16_s16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x2.nxv8i16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i16> @_Z28test_svreadz_hor_za16_s16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_hor_za16_s16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x2.nxv8i16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svreadz_hor_za16_s16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za16_s16_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i16> @test_svreadz_hor_za16_u16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_hor_za16_u16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x2.nxv8i16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i16> @_Z28test_svreadz_hor_za16_u16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_hor_za16_u16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x2.nxv8i16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svreadz_hor_za16_u16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za16_u16_vg2(1, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x half> @test_svreadz_hor_za16_f16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half> } @test_svreadz_hor_za16_f16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.horiz.x2.nxv8f16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x half> @_Z28test_svreadz_hor_za16_f16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half> } @_Z28test_svreadz_hor_za16_f16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.horiz.x2.nxv8f16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svreadz_hor_za16_f16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za16_f16_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x bfloat> @test_svreadz_hor_za16_bf16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @test_svreadz_hor_za16_bf16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.horiz.x2.nxv8bf16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x bfloat> @_Z29test_svreadz_hor_za16_bf16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @_Z29test_svreadz_hor_za16_bf16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.horiz.x2.nxv8bf16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svreadz_hor_za16_bf16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -159,150 +195,186 @@ svbfloat16x2_t test_svreadz_hor_za16_bf16_x2(uint32_t slice) __arm_streaming __a
 }
 
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i32> @test_svreadz_hor_za32_s32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_hor_za32_s32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x2.nxv4i32(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i32> @_Z28test_svreadz_hor_za32_s32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_hor_za32_s32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x2.nxv4i32(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svreadz_hor_za32_s32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za32_s32_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i32> @test_svreadz_hor_za32_u32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_hor_za32_u32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x2.nxv4i32(i32 2, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i32> @_Z28test_svreadz_hor_za32_u32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_hor_za32_u32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x2.nxv4i32(i32 2, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svreadz_hor_za32_u32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za32_u32_vg2(2, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x float> @test_svreadz_hor_za32_f32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float> } @test_svreadz_hor_za32_f32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.horiz.x2.nxv4f32(i32 3, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x float> @_Z28test_svreadz_hor_za32_f32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float> } @_Z28test_svreadz_hor_za32_f32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.horiz.x2.nxv4f32(i32 3, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svreadz_hor_za32_f32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za32_f32_vg2(3, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x i64> @test_svreadz_hor_za64_s64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_hor_za64_s64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x2.nxv2i64(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i64> @_Z28test_svreadz_hor_za64_s64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_hor_za64_s64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x2.nxv2i64(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svreadz_hor_za64_s64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za64_s64_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x i64> @test_svreadz_hor_za64_u64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_hor_za64_u64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x2.nxv2i64(i32 4, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i64> @_Z28test_svreadz_hor_za64_u64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_hor_za64_u64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x2.nxv2i64(i32 4, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svreadz_hor_za64_u64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za64_u64_vg2(4, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x double> @test_svreadz_hor_za64_f64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double> } @test_svreadz_hor_za64_f64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.horiz.x2.nxv2f64(i32 7, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x double> @_Z28test_svreadz_hor_za64_f64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double> } @_Z28test_svreadz_hor_za64_f64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.horiz.x2.nxv2f64(i32 7, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svreadz_hor_za64_f64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -314,150 +386,186 @@ svfloat64x2_t test_svreadz_hor_za64_f64_x2(uint32_t slice) __arm_streaming __arm
 // X2- ver
 //
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i8> @test_svreadz_ver_za8_s8_x2(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_ver_za8_s8_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i8> @_Z26test_svreadz_ver_za8_s8_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_ver_za8_s8_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svreadz_ver_za8_s8_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za8_s8_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i8> @test_svreadz_ver_za8_u8_x2(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_ver_za8_u8_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i8> @_Z26test_svreadz_ver_za8_u8_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_ver_za8_u8_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x2.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svreadz_ver_za8_u8_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za8_u8_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i16> @test_svreadz_ver_za16_s16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_ver_za16_s16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x2.nxv8i16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i16> @_Z28test_svreadz_ver_za16_s16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_ver_za16_s16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x2.nxv8i16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svreadz_ver_za16_s16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za16_s16_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i16> @test_svreadz_ver_za16_u16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_ver_za16_u16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x2.nxv8i16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i16> @_Z28test_svreadz_ver_za16_u16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_ver_za16_u16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x2.nxv8i16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svreadz_ver_za16_u16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za16_u16_vg2(1, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x half> @test_svreadz_ver_za16_f16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half> } @test_svreadz_ver_za16_f16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.vert.x2.nxv8f16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x half> @_Z28test_svreadz_ver_za16_f16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half> } @_Z28test_svreadz_ver_za16_f16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.vert.x2.nxv8f16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svreadz_ver_za16_f16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za16_f16_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x bfloat> @test_svreadz_ver_za16_bf16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @test_svreadz_ver_za16_bf16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.vert.x2.nxv8bf16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x bfloat> @_Z29test_svreadz_ver_za16_bf16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @_Z29test_svreadz_ver_za16_bf16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.vert.x2.nxv8bf16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svreadz_ver_za16_bf16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -465,150 +573,186 @@ svbfloat16x2_t test_svreadz_ver_za16_bf16_x2(uint32_t slice) __arm_streaming __a
 }
 
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i32> @test_svreadz_ver_za32_s32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_ver_za32_s32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x2.nxv4i32(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i32> @_Z28test_svreadz_ver_za32_s32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_ver_za32_s32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x2.nxv4i32(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svreadz_ver_za32_s32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za32_s32_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i32> @test_svreadz_ver_za32_u32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_ver_za32_u32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x2.nxv4i32(i32 2, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i32> @_Z28test_svreadz_ver_za32_u32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_ver_za32_u32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x2.nxv4i32(i32 2, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svreadz_ver_za32_u32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za32_u32_vg2(2, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x float> @test_svreadz_ver_za32_f32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float> } @test_svreadz_ver_za32_f32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.vert.x2.nxv4f32(i32 3, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x float> @_Z28test_svreadz_ver_za32_f32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float> } @_Z28test_svreadz_ver_za32_f32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.vert.x2.nxv4f32(i32 3, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svreadz_ver_za32_f32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za32_f32_vg2(3, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x i64> @test_svreadz_ver_za64_s64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_ver_za64_s64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x2.nxv2i64(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i64> @_Z28test_svreadz_ver_za64_s64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_ver_za64_s64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x2.nxv2i64(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svreadz_ver_za64_s64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za64_s64_vg2(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x i64> @test_svreadz_ver_za64_u64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_ver_za64_u64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x2.nxv2i64(i32 4, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i64> @_Z28test_svreadz_ver_za64_u64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_ver_za64_u64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x2.nxv2i64(i32 4, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svreadz_ver_za64_u64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za64_u64_vg2(4, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x double> @test_svreadz_ver_za64_f64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double> } @test_svreadz_ver_za64_f64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.vert.x2.nxv2f64(i32 7, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x double> @_Z28test_svreadz_ver_za64_f64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double> } @_Z28test_svreadz_ver_za64_f64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.vert.x2.nxv2f64(i32 7, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svreadz_ver_za64_f64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -618,9 +762,10 @@ svfloat64x2_t test_svreadz_ver_za64_f64_x2(uint32_t slice) __arm_streaming __arm
 
 //
 // X4 - hor
-// CHECK-LABEL: define dso_local <vscale x 64 x i8> @test_svreadz_hor_za8_s8_x4(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_hor_za8_s8_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -630,11 +775,14 @@ svfloat64x2_t test_svreadz_ver_za64_f64_x2(uint32_t slice) __arm_streaming __arm
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 64 x i8> @_Z26test_svreadz_hor_za8_s8_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_hor_za8_s8_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -644,16 +792,19 @@ svfloat64x2_t test_svreadz_ver_za64_f64_x2(uint32_t slice) __arm_streaming __arm
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svreadz_hor_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za8_s8_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 64 x i8> @test_svreadz_hor_za8_u8_x4(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_hor_za8_u8_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -663,11 +814,14 @@ svint8x4_t test_svreadz_hor_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inou
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 64 x i8> @_Z26test_svreadz_hor_za8_u8_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_hor_za8_u8_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.horiz.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -677,16 +831,19 @@ svint8x4_t test_svreadz_hor_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inou
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svreadz_hor_za8_u8_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za8_u8_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svreadz_hor_za16_s16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_hor_za16_s16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x4.nxv8i16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -696,11 +853,14 @@ svuint8x4_t test_svreadz_hor_za8_u8_x4(uint32_t slice) __arm_streaming __arm_ino
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z28test_svreadz_hor_za16_s16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_hor_za16_s16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x4.nxv8i16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -710,16 +870,19 @@ svuint8x4_t test_svreadz_hor_za8_u8_x4(uint32_t slice) __arm_streaming __arm_ino
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svreadz_hor_za16_s16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za16_s16_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svreadz_hor_za16_u16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_hor_za16_u16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x4.nxv8i16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -729,11 +892,14 @@ svint16x4_t test_svreadz_hor_za16_s16_x4(uint32_t slice) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z28test_svreadz_hor_za16_u16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_hor_za16_u16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.horiz.x4.nxv8i16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -743,16 +909,19 @@ svint16x4_t test_svreadz_hor_za16_s16_x4(uint32_t slice) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svreadz_hor_za16_u16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za16_u16_vg4(1, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x half> @test_svreadz_hor_za16_f16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @test_svreadz_hor_za16_f16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.horiz.x4.nxv8f16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -762,11 +931,14 @@ svuint16x4_t test_svreadz_hor_za16_u16_x4(uint32_t slice) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x half> @_Z28test_svreadz_hor_za16_f16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @_Z28test_svreadz_hor_za16_f16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.horiz.x4.nxv8f16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -776,16 +948,19 @@ svuint16x4_t test_svreadz_hor_za16_u16_x4(uint32_t slice) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svreadz_hor_za16_f16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za16_f16_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @test_svreadz_hor_za16_bf16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @test_svreadz_hor_za16_bf16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.horiz.x4.nxv8bf16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -795,11 +970,14 @@ svfloat16x4_t test_svreadz_hor_za16_f16_x4(uint32_t slice) __arm_streaming __arm
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @_Z29test_svreadz_hor_za16_bf16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @_Z29test_svreadz_hor_za16_bf16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.horiz.x4.nxv8bf16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -809,7 +987,9 @@ svfloat16x4_t test_svreadz_hor_za16_f16_x4(uint32_t slice) __arm_streaming __arm
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svreadz_hor_za16_bf16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -817,9 +997,10 @@ svbfloat16x4_t test_svreadz_hor_za16_bf16_x4(uint32_t slice) __arm_streaming __a
 }
 
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svreadz_hor_za32_s32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_hor_za32_s32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x4.nxv4i32(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -829,11 +1010,14 @@ svbfloat16x4_t test_svreadz_hor_za16_bf16_x4(uint32_t slice) __arm_streaming __a
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z28test_svreadz_hor_za32_s32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_hor_za32_s32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x4.nxv4i32(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -843,16 +1027,19 @@ svbfloat16x4_t test_svreadz_hor_za16_bf16_x4(uint32_t slice) __arm_streaming __a
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svreadz_hor_za32_s32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za32_s32_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svreadz_hor_za32_u32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_hor_za32_u32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x4.nxv4i32(i32 2, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -862,11 +1049,14 @@ svint32x4_t test_svreadz_hor_za32_s32_x4(uint32_t slice) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z28test_svreadz_hor_za32_u32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_hor_za32_u32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.horiz.x4.nxv4i32(i32 2, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -876,16 +1066,19 @@ svint32x4_t test_svreadz_hor_za32_s32_x4(uint32_t slice) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svreadz_hor_za32_u32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za32_u32_vg4(2, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x float> @test_svreadz_hor_za32_f32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @test_svreadz_hor_za32_f32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.horiz.x4.nxv4f32(i32 3, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -895,11 +1088,14 @@ svuint32x4_t test_svreadz_hor_za32_u32_x4(uint32_t slice) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x float> @_Z28test_svreadz_hor_za32_f32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @_Z28test_svreadz_hor_za32_f32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.horiz.x4.nxv4f32(i32 3, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -909,16 +1105,19 @@ svuint32x4_t test_svreadz_hor_za32_u32_x4(uint32_t slice) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svreadz_hor_za32_f32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za32_f32_vg4(3, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i64> @test_svreadz_hor_za64_s64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_hor_za64_s64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x4.nxv2i64(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -928,11 +1127,14 @@ svfloat32x4_t test_svreadz_hor_za32_f32_x4(uint32_t slice) __arm_streaming __arm
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i64> @_Z28test_svreadz_hor_za64_s64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_hor_za64_s64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x4.nxv2i64(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -942,16 +1144,19 @@ svfloat32x4_t test_svreadz_hor_za32_f32_x4(uint32_t slice) __arm_streaming __arm
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svreadz_hor_za64_s64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za64_s64_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i64> @test_svreadz_hor_za64_u64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_hor_za64_u64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x4.nxv2i64(i32 4, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -961,11 +1166,14 @@ svint64x4_t test_svreadz_hor_za64_s64_x4(uint32_t slice) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i64> @_Z28test_svreadz_hor_za64_u64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_hor_za64_u64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.horiz.x4.nxv2i64(i32 4, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -975,16 +1183,19 @@ svint64x4_t test_svreadz_hor_za64_s64_x4(uint32_t slice) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svreadz_hor_za64_u64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_hor_za64_u64_vg4(4, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x double> @test_svreadz_hor_za64_f64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @test_svreadz_hor_za64_f64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.horiz.x4.nxv2f64(i32 7, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -994,11 +1205,14 @@ svuint64x4_t test_svreadz_hor_za64_u64_x4(uint32_t slice) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x double> @_Z28test_svreadz_hor_za64_f64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @_Z28test_svreadz_hor_za64_f64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.horiz.x4.nxv2f64(i32 7, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1008,7 +1222,9 @@ svuint64x4_t test_svreadz_hor_za64_u64_x4(uint32_t slice) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svreadz_hor_za64_f64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -1017,9 +1233,10 @@ svfloat64x4_t test_svreadz_hor_za64_f64_x4(uint32_t slice) __arm_streaming __arm
 
 //
 // X4 - ver
-// CHECK-LABEL: define dso_local <vscale x 64 x i8> @test_svreadz_ver_za8_s8_x4(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_ver_za8_s8_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1029,11 +1246,14 @@ svfloat64x4_t test_svreadz_hor_za64_f64_x4(uint32_t slice) __arm_streaming __arm
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 64 x i8> @_Z26test_svreadz_ver_za8_s8_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_ver_za8_s8_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1043,16 +1263,19 @@ svfloat64x4_t test_svreadz_hor_za64_f64_x4(uint32_t slice) __arm_streaming __arm
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svreadz_ver_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za8_s8_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 64 x i8> @test_svreadz_ver_za8_u8_x4(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_ver_za8_u8_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1062,11 +1285,14 @@ svint8x4_t test_svreadz_ver_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inou
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 64 x i8> @_Z26test_svreadz_ver_za8_u8_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z26test_svreadz_ver_za8_u8_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.vert.x4.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -1076,16 +1302,19 @@ svint8x4_t test_svreadz_ver_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inou
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svreadz_ver_za8_u8_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za8_u8_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svreadz_ver_za16_s16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_ver_za16_s16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x4.nxv8i16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1095,11 +1324,14 @@ svuint8x4_t test_svreadz_ver_za8_u8_x4(uint32_t slice) __arm_streaming __arm_ino
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z28test_svreadz_ver_za16_s16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_ver_za16_s16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x4.nxv8i16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1109,16 +1341,19 @@ svuint8x4_t test_svreadz_ver_za8_u8_x4(uint32_t slice) __arm_streaming __arm_ino
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svreadz_ver_za16_s16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za16_s16_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svreadz_ver_za16_u16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_ver_za16_u16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x4.nxv8i16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1128,11 +1363,14 @@ svint16x4_t test_svreadz_ver_za16_s16_x4(uint32_t slice) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z28test_svreadz_ver_za16_u16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z28test_svreadz_ver_za16_u16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.vert.x4.nxv8i16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -1142,16 +1380,19 @@ svint16x4_t test_svreadz_ver_za16_s16_x4(uint32_t slice) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svreadz_ver_za16_u16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za16_u16_vg4(1, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x half> @test_svreadz_ver_za16_f16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @test_svreadz_ver_za16_f16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.vert.x4.nxv8f16(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -1161,11 +1402,14 @@ svuint16x4_t test_svreadz_ver_za16_u16_x4(uint32_t slice) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x half> @_Z28test_svreadz_ver_za16_f16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @_Z28test_svreadz_ver_za16_f16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.vert.x4.nxv8f16(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -1175,16 +1419,19 @@ svuint16x4_t test_svreadz_ver_za16_u16_x4(uint32_t slice) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svreadz_ver_za16_f16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za16_f16_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @test_svreadz_ver_za16_bf16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @test_svreadz_ver_za16_bf16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.vert.x4.nxv8bf16(i32 1, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -1194,11 +1441,14 @@ svfloat16x4_t test_svreadz_ver_za16_f16_x4(uint32_t slice) __arm_streaming __arm
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @_Z29test_svreadz_ver_za16_bf16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @_Z29test_svreadz_ver_za16_bf16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.vert.x4.nxv8bf16(i32 1, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -1208,7 +1458,9 @@ svfloat16x4_t test_svreadz_ver_za16_f16_x4(uint32_t slice) __arm_streaming __arm
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svreadz_ver_za16_bf16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -1216,9 +1468,10 @@ svbfloat16x4_t test_svreadz_ver_za16_bf16_x4(uint32_t slice) __arm_streaming __a
 }
 
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svreadz_ver_za32_s32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_ver_za32_s32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x4.nxv4i32(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1228,11 +1481,14 @@ svbfloat16x4_t test_svreadz_ver_za16_bf16_x4(uint32_t slice) __arm_streaming __a
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z28test_svreadz_ver_za32_s32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_ver_za32_s32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x4.nxv4i32(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1242,16 +1498,19 @@ svbfloat16x4_t test_svreadz_ver_za16_bf16_x4(uint32_t slice) __arm_streaming __a
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svreadz_ver_za32_s32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za32_s32_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svreadz_ver_za32_u32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_ver_za32_u32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x4.nxv4i32(i32 2, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1261,11 +1520,14 @@ svint32x4_t test_svreadz_ver_za32_s32_x4(uint32_t slice) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z28test_svreadz_ver_za32_u32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z28test_svreadz_ver_za32_u32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.vert.x4.nxv4i32(i32 2, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -1275,16 +1537,19 @@ svint32x4_t test_svreadz_ver_za32_s32_x4(uint32_t slice) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svreadz_ver_za32_u32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za32_u32_vg4(2, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x float> @test_svreadz_ver_za32_f32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @test_svreadz_ver_za32_f32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.vert.x4.nxv4f32(i32 3, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -1294,11 +1559,14 @@ svuint32x4_t test_svreadz_ver_za32_u32_x4(uint32_t slice) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x float> @_Z28test_svreadz_ver_za32_f32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @_Z28test_svreadz_ver_za32_f32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.vert.x4.nxv4f32(i32 3, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -1308,16 +1576,19 @@ svuint32x4_t test_svreadz_ver_za32_u32_x4(uint32_t slice) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svreadz_ver_za32_f32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za32_f32_vg4(3, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i64> @test_svreadz_ver_za64_s64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_ver_za64_s64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x4.nxv2i64(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1327,11 +1598,14 @@ svfloat32x4_t test_svreadz_ver_za32_f32_x4(uint32_t slice) __arm_streaming __arm
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i64> @_Z28test_svreadz_ver_za64_s64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_ver_za64_s64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x4.nxv2i64(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1341,16 +1615,19 @@ svfloat32x4_t test_svreadz_ver_za32_f32_x4(uint32_t slice) __arm_streaming __arm
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svreadz_ver_za64_s64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za64_s64_vg4(0, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i64> @test_svreadz_ver_za64_u64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_ver_za64_u64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x4.nxv2i64(i32 4, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1360,11 +1637,14 @@ svint64x4_t test_svreadz_ver_za64_s64_x4(uint32_t slice) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i64> @_Z28test_svreadz_ver_za64_u64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z28test_svreadz_ver_za64_u64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.vert.x4.nxv2i64(i32 4, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -1374,16 +1654,19 @@ svint64x4_t test_svreadz_ver_za64_s64_x4(uint32_t slice) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svreadz_ver_za64_u64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_ver_za64_u64_vg4(4, slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x double> @test_svreadz_ver_za64_f64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @test_svreadz_ver_za64_f64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.vert.x4.nxv2f64(i32 7, i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1393,11 +1676,14 @@ svuint64x4_t test_svreadz_ver_za64_u64_x4(uint32_t slice) __arm_streaming __arm_
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x double> @_Z28test_svreadz_ver_za64_f64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @_Z28test_svreadz_ver_za64_f64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.vert.x4.nxv2f64(i32 7, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -1407,7 +1693,9 @@ svuint64x4_t test_svreadz_ver_za64_u64_x4(uint32_t slice) __arm_streaming __arm_
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svreadz_ver_za64_f64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -1415,13 +1703,13 @@ svfloat64x4_t test_svreadz_ver_za64_f64_x4(uint32_t slice) __arm_streaming __arm
 }
 
 // CHECK-LABEL: define dso_local <vscale x 16 x i8> @test_svreadz_hor_za8_s8(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sme.readz.horiz.nxv16i8(i32 0, i32 [[SLICE]])
 // CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
 // CPP-CHECK-LABEL: define dso_local <vscale x 16 x i8> @_Z23test_svreadz_hor_za8_s8j(
-// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0:[0-9]+]] {
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sme.readz.horiz.nxv16i8(i32 0, i32 [[SLICE]])
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
@@ -1829,300 +2117,372 @@ svfloat64_t test_svreadz_hor_za128_f64(uint32_t slice) __arm_streaming __arm_ino
 // X2
 //
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i8> @test_svreadz_za8_s8_x2(
-// CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_za8_s8_x2(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x2.nxv16i8(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i8> @_Z22test_svreadz_za8_s8_x2j(
-// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0:[0-9]+]] {
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z22test_svreadz_za8_s8_x2j(
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x2.nxv16i8(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svint8x2_t test_svreadz_za8_s8_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za8_s8_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i8> @test_svreadz_za8_u8_x2(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_za8_u8_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x2.nxv16i8(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i8> @_Z22test_svreadz_za8_u8_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z22test_svreadz_za8_u8_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x2.nxv16i8(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i8> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP5]]
 //
 svuint8x2_t test_svreadz_za8_u8_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za8_u8_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i16> @test_svreadz_za16_s16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_za16_s16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x2.nxv8i16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i16> @_Z24test_svreadz_za16_s16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z24test_svreadz_za16_s16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x2.nxv8i16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svint16x2_t test_svreadz_za16_s16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_s16_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i16> @test_svreadz_za16_u16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_za16_u16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x2.nxv8i16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i16> @_Z24test_svreadz_za16_u16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z24test_svreadz_za16_u16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x2.nxv8i16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i16> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]]
 //
 svuint16x2_t test_svreadz_za16_u16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_u16_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i32> @test_svreadz_za32_s32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_za32_s32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x2.nxv4i32(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i32> @_Z24test_svreadz_za32_s32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z24test_svreadz_za32_s32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x2.nxv4i32(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svint32x2_t test_svreadz_za32_s32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za32_s32_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i32> @test_svreadz_za32_u32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_za32_u32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x2.nxv4i32(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i32> @_Z24test_svreadz_za32_u32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z24test_svreadz_za32_u32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x2.nxv4i32(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i32> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]]
 //
 svuint32x2_t test_svreadz_za32_u32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za32_u32_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x i64> @test_svreadz_za64_s64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_za64_s64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x2.nxv2i64(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i64> @_Z24test_svreadz_za64_s64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z24test_svreadz_za64_s64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x2.nxv2i64(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svint64x2_t test_svreadz_za64_s64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za64_s64_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x i64> @test_svreadz_za64_u64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_za64_u64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x2.nxv2i64(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i64> @_Z24test_svreadz_za64_u64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z24test_svreadz_za64_u64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x2.nxv2i64(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x i64> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]]
 //
 svuint64x2_t test_svreadz_za64_u64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za64_u64_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x bfloat> @test_svreadz_za16_bf16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @test_svreadz_za16_bf16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.x2.nxv8bf16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x bfloat> @_Z25test_svreadz_za16_bf16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @_Z25test_svreadz_za16_bf16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.x2.nxv8bf16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x bfloat> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP5]]
 //
 svbfloat16x2_t test_svreadz_za16_bf16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_bf16_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x half> @test_svreadz_za16_f16_x2(
+// CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half> } @test_svreadz_za16_f16_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.x2.nxv8f16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x half> @_Z24test_svreadz_za16_f16_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half> } @_Z24test_svreadz_za16_f16_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.x2.nxv8f16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x half> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP5]]
 //
 svfloat16x2_t test_svreadz_za16_f16_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_f16_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x float> @test_svreadz_za32_f32_x2(
+// CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float> } @test_svreadz_za32_f32_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.x2.nxv4f32(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x float> @_Z24test_svreadz_za32_f32_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float> } @_Z24test_svreadz_za32_f32_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.x2.nxv4f32(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x float> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP5]]
 //
 svfloat32x2_t test_svreadz_za32_f32_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za32_f32_vg1x2(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 4 x double> @test_svreadz_za64_f64_x2(
+// CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double> } @test_svreadz_za64_f64_x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.x2.nxv2f64(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 4 x double> @_Z24test_svreadz_za64_f64_x2j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double> } @_Z24test_svreadz_za64_f64_x2j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.x2.nxv2f64(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
 // CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 1
 // CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP4]]
+// CPP-CHECK-NEXT:    store <vscale x 4 x double> [[TMP4]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP5]]
 //
 svfloat64x2_t test_svreadz_za64_f64_x2(uint32_t slice) __arm_streaming __arm_inout("za")
 {
@@ -2133,9 +2493,10 @@ svfloat64x2_t test_svreadz_za64_f64_x2(uint32_t slice) __arm_streaming __arm_ino
 // X4
 //
 
-// CHECK-LABEL: define dso_local <vscale x 64 x i8> @test_svreadz_za8_s8_x4(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_za8_s8_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x4.nxv16i8(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -2145,11 +2506,14 @@ svfloat64x2_t test_svreadz_za64_f64_x2(uint32_t slice) __arm_streaming __arm_ino
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 64 x i8> @_Z22test_svreadz_za8_s8_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z22test_svreadz_za8_s8_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x4.nxv16i8(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -2159,16 +2523,19 @@ svfloat64x2_t test_svreadz_za64_f64_x2(uint32_t slice) __arm_streaming __arm_ino
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svint8x4_t test_svreadz_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za8_s8_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 64 x i8> @test_svreadz_za8_u8_x4(
+// CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @test_svreadz_za8_u8_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x4.nxv16i8(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -2178,11 +2545,14 @@ svint8x4_t test_svreadz_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inout("z
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 64 x i8> @_Z22test_svreadz_za8_u8_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @_Z22test_svreadz_za8_u8_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sme.readz.x4.nxv16i8(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
@@ -2192,16 +2562,19 @@ svint8x4_t test_svreadz_za8_s8_x4(uint32_t slice) __arm_streaming __arm_inout("z
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 64 x i8> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP9]]
 //
 svuint8x4_t test_svreadz_za8_u8_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za8_u8_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svreadz_za16_s16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_za16_s16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x4.nxv8i16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -2211,11 +2584,14 @@ svuint8x4_t test_svreadz_za8_u8_x4(uint32_t slice) __arm_streaming __arm_inout("
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z24test_svreadz_za16_s16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z24test_svreadz_za16_s16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x4.nxv8i16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -2225,16 +2601,19 @@ svuint8x4_t test_svreadz_za8_u8_x4(uint32_t slice) __arm_streaming __arm_inout("
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svint16x4_t test_svreadz_za16_s16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_s16_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x i16> @test_svreadz_za16_u16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @test_svreadz_za16_u16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x4.nxv8i16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -2244,11 +2623,14 @@ svint16x4_t test_svreadz_za16_s16_x4(uint32_t slice) __arm_streaming __arm_inout
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x i16> @_Z24test_svreadz_za16_u16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @_Z24test_svreadz_za16_u16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sme.readz.x4.nxv8i16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
@@ -2258,16 +2640,19 @@ svint16x4_t test_svreadz_za16_s16_x4(uint32_t slice) __arm_streaming __arm_inout
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x i16> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP9]]
 //
 svuint16x4_t test_svreadz_za16_u16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_u16_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svreadz_za32_s32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_za32_s32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x4.nxv4i32(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -2277,11 +2662,14 @@ svuint16x4_t test_svreadz_za16_u16_x4(uint32_t slice) __arm_streaming __arm_inou
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z24test_svreadz_za32_s32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z24test_svreadz_za32_s32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x4.nxv4i32(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -2291,16 +2679,19 @@ svuint16x4_t test_svreadz_za16_u16_x4(uint32_t slice) __arm_streaming __arm_inou
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svint32x4_t test_svreadz_za32_s32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za32_s32_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x i32> @test_svreadz_za32_u32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @test_svreadz_za32_u32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x4.nxv4i32(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -2310,11 +2701,14 @@ svint32x4_t test_svreadz_za32_s32_x4(uint32_t slice) __arm_streaming __arm_inout
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i32> @_Z24test_svreadz_za32_u32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @_Z24test_svreadz_za32_u32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sme.readz.x4.nxv4i32(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
@@ -2324,16 +2718,19 @@ svint32x4_t test_svreadz_za32_s32_x4(uint32_t slice) __arm_streaming __arm_inout
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x i32> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP9]]
 //
 svuint32x4_t test_svreadz_za32_u32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za32_u32_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i64> @test_svreadz_za64_s64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_za64_s64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x4.nxv2i64(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -2343,11 +2740,14 @@ svuint32x4_t test_svreadz_za32_u32_x4(uint32_t slice) __arm_streaming __arm_inou
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i64> @_Z24test_svreadz_za64_s64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z24test_svreadz_za64_s64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x4.nxv2i64(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -2357,16 +2757,19 @@ svuint32x4_t test_svreadz_za32_u32_x4(uint32_t slice) __arm_streaming __arm_inou
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svint64x4_t test_svreadz_za64_s64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za64_s64_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x i64> @test_svreadz_za64_u64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @test_svreadz_za64_u64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x4.nxv2i64(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -2376,11 +2779,14 @@ svint64x4_t test_svreadz_za64_s64_x4(uint32_t slice) __arm_streaming __arm_inout
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i64> @_Z24test_svreadz_za64_u64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @_Z24test_svreadz_za64_u64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sme.readz.x4.nxv2i64(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
@@ -2390,16 +2796,19 @@ svint64x4_t test_svreadz_za64_s64_x4(uint32_t slice) __arm_streaming __arm_inout
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x i64> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP9]]
 //
 svuint64x4_t test_svreadz_za64_u64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za64_u64_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @test_svreadz_za16_bf16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @test_svreadz_za16_bf16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.x4.nxv8bf16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -2409,11 +2818,14 @@ svuint64x4_t test_svreadz_za64_u64_x4(uint32_t slice) __arm_streaming __arm_inou
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x bfloat> @_Z25test_svreadz_za16_bf16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @_Z25test_svreadz_za16_bf16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sme.readz.x4.nxv8bf16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP1]], i64 0)
@@ -2423,16 +2835,19 @@ svuint64x4_t test_svreadz_za64_u64_x4(uint32_t slice) __arm_streaming __arm_inou
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x bfloat> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP9]]
 //
 svbfloat16x4_t test_svreadz_za16_bf16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_bf16_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 32 x half> @test_svreadz_za16_f16_x4(
+// CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @test_svreadz_za16_f16_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.x4.nxv8f16(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -2442,11 +2857,14 @@ svbfloat16x4_t test_svreadz_za16_bf16_x4(uint32_t slice) __arm_streaming __arm_i
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 32 x half> @_Z24test_svreadz_za16_f16_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @_Z24test_svreadz_za16_f16_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.readz.x4.nxv8f16(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP1]], i64 0)
@@ -2456,16 +2874,19 @@ svbfloat16x4_t test_svreadz_za16_bf16_x4(uint32_t slice) __arm_streaming __arm_i
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 16)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 32 x half> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP9]]
 //
 svfloat16x4_t test_svreadz_za16_f16_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za16_f16_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 16 x float> @test_svreadz_za32_f32_x4(
+// CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @test_svreadz_za32_f32_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.x4.nxv4f32(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -2475,11 +2896,14 @@ svfloat16x4_t test_svreadz_za16_f16_x4(uint32_t slice) __arm_streaming __arm_ino
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 16 x float> @_Z24test_svreadz_za32_f32_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @_Z24test_svreadz_za32_f32_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.readz.x4.nxv4f32(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP1]], i64 0)
@@ -2489,16 +2913,19 @@ svfloat16x4_t test_svreadz_za16_f16_x4(uint32_t slice) __arm_streaming __arm_ino
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 8)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 16 x float> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP9]]
 //
 svfloat32x4_t test_svreadz_za32_f32_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {
    return svreadz_za32_f32_vg1x4(slice);
 }
 
-// CHECK-LABEL: define dso_local <vscale x 8 x double> @test_svreadz_za64_f64_x4(
+// CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @test_svreadz_za64_f64_x4(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.x4.nxv2f64(i32 [[SLICE]])
 // CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -2508,11 +2935,14 @@ svfloat32x4_t test_svreadz_za32_f32_x4(uint32_t slice) __arm_streaming __arm_ino
 // CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
-// CPP-CHECK-LABEL: define dso_local <vscale x 8 x double> @_Z24test_svreadz_za64_f64_x4j(
+// CPP-CHECK-LABEL: define dso_local { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @_Z24test_svreadz_za64_f64_x4j(
 // CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]]) #[[ATTR0]] {
 // CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[RETVAL:%.*]] = alloca { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, align 16
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.readz.x4.nxv2f64(i32 [[SLICE]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 0
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP1]], i64 0)
@@ -2522,7 +2952,9 @@ svfloat32x4_t test_svreadz_za32_f32_x4(uint32_t slice) __arm_streaming __arm_ino
 // CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 4)
 // CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], 3
 // CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    store <vscale x 8 x double> [[TMP8]], ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = load { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }, ptr [[RETVAL]], align 16
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP9]]
 //
 svfloat64x4_t test_svreadz_za64_f64_x4(uint32_t slice) __arm_streaming __arm_inout("za")
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2-bfloat.c
index abcb48fb8fa1a0..4691172b14a69f 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2-bfloat.c
@@ -24,15 +24,15 @@
 
 // CHECK-LABEL: @test_svcreate2_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svcreate2_bf16u14__SVBfloat16_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 svbfloat16x2_t test_svcreate2_bf16(svbfloat16_t x0, svbfloat16_t x1) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2.c
index 2338f80fd08966..275908eb819c94 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create2.c
@@ -24,15 +24,15 @@
 
 // CHECK-LABEL: @test_svcreate2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svcreate2_s8u10__SVInt8_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 svint8x2_t test_svcreate2_s8(svint8_t x0, svint8_t x1) ATTR
 {
@@ -41,15 +41,15 @@ svint8x2_t test_svcreate2_s8(svint8_t x0, svint8_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_s16u11__SVInt16_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 svint16x2_t test_svcreate2_s16(svint16_t x0, svint16_t x1) ATTR
 {
@@ -58,15 +58,15 @@ svint16x2_t test_svcreate2_s16(svint16_t x0, svint16_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_s32u11__SVInt32_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 svint32x2_t test_svcreate2_s32(svint32_t x0, svint32_t x1) ATTR
 {
@@ -75,15 +75,15 @@ svint32x2_t test_svcreate2_s32(svint32_t x0, svint32_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_s64u11__SVInt64_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 svint64x2_t test_svcreate2_s64(svint64_t x0, svint64_t x1) ATTR
 {
@@ -92,15 +92,15 @@ svint64x2_t test_svcreate2_s64(svint64_t x0, svint64_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svcreate2_u8u11__SVUint8_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 svuint8x2_t test_svcreate2_u8(svuint8_t x0, svuint8_t x1) ATTR
 {
@@ -109,15 +109,15 @@ svuint8x2_t test_svcreate2_u8(svuint8_t x0, svuint8_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_u16u12__SVUint16_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 svuint16x2_t test_svcreate2_u16(svuint16_t x0, svuint16_t x1) ATTR
 {
@@ -126,15 +126,15 @@ svuint16x2_t test_svcreate2_u16(svuint16_t x0, svuint16_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_u32u12__SVUint32_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 svuint32x2_t test_svcreate2_u32(svuint32_t x0, svuint32_t x1) ATTR
 {
@@ -143,15 +143,15 @@ svuint32x2_t test_svcreate2_u32(svuint32_t x0, svuint32_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_u64u12__SVUint64_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 svuint64x2_t test_svcreate2_u64(svuint64_t x0, svuint64_t x1) ATTR
 {
@@ -160,15 +160,15 @@ svuint64x2_t test_svcreate2_u64(svuint64_t x0, svuint64_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP0]], <vscale x 8 x half> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_f16u13__SVFloat16_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP0]], <vscale x 8 x half> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]]
 //
 svfloat16x2_t test_svcreate2_f16(svfloat16_t x0, svfloat16_t x1) ATTR
 {
@@ -177,15 +177,15 @@ svfloat16x2_t test_svcreate2_f16(svfloat16_t x0, svfloat16_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP0]], <vscale x 4 x float> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_f32u13__SVFloat32_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP0]], <vscale x 4 x float> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]]
 //
 svfloat32x2_t test_svcreate2_f32(svfloat32_t x0, svfloat32_t x1) ATTR
 {
@@ -194,15 +194,15 @@ svfloat32x2_t test_svcreate2_f32(svfloat32_t x0, svfloat32_t x1) ATTR
 
 // CHECK-LABEL: @test_svcreate2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP0]], <vscale x 2 x double> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP1]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[X1:%.*]], 1
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate2_f64u13__SVFloat64_tS_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP0]], <vscale x 2 x double> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP1]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]]
 //
 svfloat64x2_t test_svcreate2_f64(svfloat64_t x0, svfloat64_t x1) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3-bfloat.c
index 86533e58f56178..3e2bd259e5c7c1 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3-bfloat.c
@@ -24,17 +24,17 @@
 
 // CHECK-LABEL: @test_svcreate3_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svcreate3_bf16u14__SVBfloat16_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 svbfloat16x3_t test_svcreate3_bf16(svbfloat16_t x0, svbfloat16_t x1, svbfloat16_t x2) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3.c
index aeff07104c189f..1395d92cc06ce0 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create3.c
@@ -24,17 +24,17 @@
 
 // CHECK-LABEL: @test_svcreate3_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svcreate3_s8u10__SVInt8_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]]
 //
 svint8x3_t test_svcreate3_s8(svint8_t x0, svint8_t x1, svint8_t x2) ATTR
 {
@@ -43,17 +43,17 @@ svint8x3_t test_svcreate3_s8(svint8_t x0, svint8_t x1, svint8_t x2) ATTR
 
 // CHECK-LABEL: @test_svcreate3_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_s16u11__SVInt16_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 svint16x3_t test_svcreate3_s16(svint16_t x0, svint16_t x1, svint16_t x2) ATTR
 {
@@ -62,17 +62,17 @@ svint16x3_t test_svcreate3_s16(svint16_t x0, svint16_t x1, svint16_t x2) ATTR
 
 // CHECK-LABEL: @test_svcreate3_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_s32u11__SVInt32_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 svint32x3_t test_svcreate3_s32(svint32_t x0, svint32_t x1, svint32_t x2) ATTR
 {
@@ -81,17 +81,17 @@ svint32x3_t test_svcreate3_s32(svint32_t x0, svint32_t x1, svint32_t x2) ATTR
 
 // CHECK-LABEL: @test_svcreate3_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_s64u11__SVInt64_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 svint64x3_t test_svcreate3_s64(svint64_t x0, svint64_t x1, svint64_t x2) ATTR
 {
@@ -100,17 +100,17 @@ svint64x3_t test_svcreate3_s64(svint64_t x0, svint64_t x1, svint64_t x2) ATTR
 
 // CHECK-LABEL: @test_svcreate3_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svcreate3_u8u11__SVUint8_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]]
 //
 svuint8x3_t test_svcreate3_u8(svuint8_t x0, svuint8_t x1, svuint8_t x2) ATTR
 {
@@ -119,17 +119,17 @@ svuint8x3_t test_svcreate3_u8(svuint8_t x0, svuint8_t x1, svuint8_t x2) ATTR
 
 // CHECK-LABEL: @test_svcreate3_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_u16u12__SVUint16_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 svuint16x3_t test_svcreate3_u16(svuint16_t x0, svuint16_t x1, svuint16_t x2) ATTR
 {
@@ -138,17 +138,17 @@ svuint16x3_t test_svcreate3_u16(svuint16_t x0, svuint16_t x1, svuint16_t x2) ATT
 
 // CHECK-LABEL: @test_svcreate3_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_u32u12__SVUint32_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 svuint32x3_t test_svcreate3_u32(svuint32_t x0, svuint32_t x1, svuint32_t x2) ATTR
 {
@@ -157,17 +157,17 @@ svuint32x3_t test_svcreate3_u32(svuint32_t x0, svuint32_t x1, svuint32_t x2) ATT
 
 // CHECK-LABEL: @test_svcreate3_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_u64u12__SVUint64_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 svuint64x3_t test_svcreate3_u64(svuint64_t x0, svuint64_t x1, svuint64_t x2) ATTR
 {
@@ -176,17 +176,17 @@ svuint64x3_t test_svcreate3_u64(svuint64_t x0, svuint64_t x1, svuint64_t x2) ATT
 
 // CHECK-LABEL: @test_svcreate3_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP0]], <vscale x 8 x half> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP1]], <vscale x 8 x half> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x half> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_f16u13__SVFloat16_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP0]], <vscale x 8 x half> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP1]], <vscale x 8 x half> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x half> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]]
 //
 svfloat16x3_t test_svcreate3_f16(svfloat16_t x0, svfloat16_t x1, svfloat16_t x2) ATTR
 {
@@ -195,17 +195,17 @@ svfloat16x3_t test_svcreate3_f16(svfloat16_t x0, svfloat16_t x1, svfloat16_t x2)
 
 // CHECK-LABEL: @test_svcreate3_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP0]], <vscale x 4 x float> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP1]], <vscale x 4 x float> [[X2:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x float> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_f32u13__SVFloat32_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP0]], <vscale x 4 x float> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP1]], <vscale x 4 x float> [[X2:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x float> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]]
 //
 svfloat32x3_t test_svcreate3_f32(svfloat32_t x0, svfloat32_t x1, svfloat32_t x2) ATTR
 {
@@ -214,17 +214,17 @@ svfloat32x3_t test_svcreate3_f32(svfloat32_t x0, svfloat32_t x1, svfloat32_t x2)
 
 // CHECK-LABEL: @test_svcreate3_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP0]], <vscale x 2 x double> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP1]], <vscale x 2 x double> [[X2:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x double> [[TMP2]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[X2:%.*]], 2
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate3_f64u13__SVFloat64_tS_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP0]], <vscale x 2 x double> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP1]], <vscale x 2 x double> [[X2:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x double> [[TMP2]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]]
 //
 svfloat64x3_t test_svcreate3_f64(svfloat64_t x0, svfloat64_t x1, svfloat64_t x2) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4-bfloat.c
index 3067ae4875719c..e821c7501d7a9b 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4-bfloat.c
@@ -24,19 +24,19 @@
 
 // CHECK-LABEL: @test_svcreate4_bf16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[X4:%.*]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svcreate4_bf16u14__SVBfloat16_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[X4:%.*]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]]
 //
 svbfloat16x4_t test_svcreate4_bf16(svbfloat16_t x0, svbfloat16_t x1, svbfloat16_t x2, svbfloat16_t x4) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4.c
index b2781e97f7ec2e..97672e2833bebf 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_create4.c
@@ -24,19 +24,19 @@
 
 // CHECK-LABEL: @test_svcreate4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svcreate4_s8u10__SVInt8_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]]
 //
 svint8x4_t test_svcreate4_s8(svint8_t x0, svint8_t x1, svint8_t x2, svint8_t x4) ATTR
 {
@@ -45,19 +45,19 @@ svint8x4_t test_svcreate4_s8(svint8_t x0, svint8_t x1, svint8_t x2, svint8_t x4)
 
 // CHECK-LABEL: @test_svcreate4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_s16u11__SVInt16_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]]
 //
 svint16x4_t test_svcreate4_s16(svint16_t x0, svint16_t x1, svint16_t x2, svint16_t x4) ATTR
 {
@@ -66,19 +66,19 @@ svint16x4_t test_svcreate4_s16(svint16_t x0, svint16_t x1, svint16_t x2, svint16
 
 // CHECK-LABEL: @test_svcreate4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_s32u11__SVInt32_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]]
 //
 svint32x4_t test_svcreate4_s32(svint32_t x0, svint32_t x1, svint32_t x2, svint32_t x4) ATTR
 {
@@ -87,19 +87,19 @@ svint32x4_t test_svcreate4_s32(svint32_t x0, svint32_t x1, svint32_t x2, svint32
 
 // CHECK-LABEL: @test_svcreate4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_s64u11__SVInt64_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]]
 //
 svint64x4_t test_svcreate4_s64(svint64_t x0, svint64_t x1, svint64_t x2, svint64_t x4) ATTR
 {
@@ -108,19 +108,19 @@ svint64x4_t test_svcreate4_s64(svint64_t x0, svint64_t x1, svint64_t x2, svint64
 
 // CHECK-LABEL: @test_svcreate4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z17test_svcreate4_u8u11__SVUint8_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]]
 //
 svuint8x4_t test_svcreate4_u8(svuint8_t x0, svuint8_t x1, svuint8_t x2, svuint8_t x4) ATTR
 {
@@ -129,19 +129,19 @@ svuint8x4_t test_svcreate4_u8(svuint8_t x0, svuint8_t x1, svuint8_t x2, svuint8_
 
 // CHECK-LABEL: @test_svcreate4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_u16u12__SVUint16_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]]
 //
 svuint16x4_t test_svcreate4_u16(svuint16_t x0, svuint16_t x1, svuint16_t x2, svuint16_t x4) ATTR
 {
@@ -150,19 +150,19 @@ svuint16x4_t test_svcreate4_u16(svuint16_t x0, svuint16_t x1, svuint16_t x2, svu
 
 // CHECK-LABEL: @test_svcreate4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_u32u12__SVUint32_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i32> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]]
 //
 svuint32x4_t test_svcreate4_u32(svuint32_t x0, svuint32_t x1, svuint32_t x2, svuint32_t x4) ATTR
 {
@@ -171,19 +171,19 @@ svuint32x4_t test_svcreate4_u32(svuint32_t x0, svuint32_t x1, svuint32_t x2, svu
 
 // CHECK-LABEL: @test_svcreate4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_u64u12__SVUint64_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i64> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]]
 //
 svuint64x4_t test_svcreate4_u64(svuint64_t x0, svuint64_t x1, svuint64_t x2, svuint64_t x4) ATTR
 {
@@ -192,19 +192,19 @@ svuint64x4_t test_svcreate4_u64(svuint64_t x0, svuint64_t x1, svuint64_t x2, svu
 
 // CHECK-LABEL: @test_svcreate4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP0]], <vscale x 8 x half> [[X1:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP1]], <vscale x 8 x half> [[X2:%.*]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[X4:%.*]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_f16u13__SVFloat16_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP0]], <vscale x 8 x half> [[X1:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP1]], <vscale x 8 x half> [[X2:%.*]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP2]], <vscale x 8 x half> [[X4:%.*]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]]
 //
 svfloat16x4_t test_svcreate4_f16(svfloat16_t x0, svfloat16_t x1, svfloat16_t x2, svfloat16_t x4) ATTR
 {
@@ -213,19 +213,19 @@ svfloat16x4_t test_svcreate4_f16(svfloat16_t x0, svfloat16_t x1, svfloat16_t x2,
 
 // CHECK-LABEL: @test_svcreate4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP0]], <vscale x 4 x float> [[X1:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP1]], <vscale x 4 x float> [[X2:%.*]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[X4:%.*]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_f32u13__SVFloat32_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP0]], <vscale x 4 x float> [[X1:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP1]], <vscale x 4 x float> [[X2:%.*]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP2]], <vscale x 4 x float> [[X4:%.*]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]]
 //
 svfloat32x4_t test_svcreate4_f32(svfloat32_t x0, svfloat32_t x1, svfloat32_t x2, svfloat32_t x4) ATTR
 {
@@ -234,19 +234,19 @@ svfloat32x4_t test_svcreate4_f32(svfloat32_t x0, svfloat32_t x1, svfloat32_t x2,
 
 // CHECK-LABEL: @test_svcreate4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[X0:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP0]], <vscale x 2 x double> [[X1:%.*]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP1]], <vscale x 2 x double> [[X2:%.*]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[X4:%.*]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP3]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[X0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[X1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[X2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[X4:%.*]], 3
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svcreate4_f64u13__SVFloat64_tS_S_S_(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[X0:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP0]], <vscale x 2 x double> [[X1:%.*]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP1]], <vscale x 2 x double> [[X2:%.*]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP2]], <vscale x 2 x double> [[X4:%.*]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP3]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[X0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[X1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[X2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[X4:%.*]], 3
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]]
 //
 svfloat64x4_t test_svcreate4_f64(svfloat64_t x0, svfloat64_t x1, svfloat64_t x2, svfloat64_t x4) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2-bfloat.c
index eb66fbf0558697..05d9ec66847944 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2-bfloat.c
@@ -24,13 +24,17 @@
 
 // CHECK-LABEL: @test_svget2_bf16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget2_bf16_014svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
 //
 svbfloat16_t test_svget2_bf16_0(svbfloat16x2_t tuple) ATTR
 {
@@ -39,13 +43,17 @@ svbfloat16_t test_svget2_bf16_0(svbfloat16x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_bf16_1(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget2_bf16_114svbfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
 //
 svbfloat16_t test_svget2_bf16_1(svbfloat16x2_t tuple) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2.c
index 6f1b0e415bfecb..c9693583f2f539 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get2.c
@@ -25,13 +25,17 @@
 
 // CHECK-LABEL: @test_svget2_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svget2_s810svint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP2]]
 //
 svint8_t test_svget2_s8(svint8x2_t tuple) ATTR
 {
@@ -40,13 +44,17 @@ svint8_t test_svget2_s8(svint8x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_s1611svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 svint16_t test_svget2_s16(svint16x2_t tuple) ATTR
 {
@@ -55,13 +63,17 @@ svint16_t test_svget2_s16(svint16x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_s3211svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 svint32_t test_svget2_s32(svint32x2_t tuple) ATTR
 {
@@ -70,13 +82,17 @@ svint32_t test_svget2_s32(svint32x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[TUPLE:%.*]], i64 2)
-// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_s6411svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[TUPLE:%.*]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 svint64_t test_svget2_s64(svint64x2_t tuple) ATTR
 {
@@ -85,13 +101,17 @@ svint64_t test_svget2_s64(svint64x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svget2_u811svuint8x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv32i8(<vscale x 32 x i8> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP2]]
 //
 svuint8_t test_svget2_u8(svuint8x2_t tuple) ATTR
 {
@@ -100,13 +120,17 @@ svuint8_t test_svget2_u8(svuint8x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_u1612svuint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP2]]
 //
 svuint16_t test_svget2_u16(svuint16x2_t tuple) ATTR
 {
@@ -115,13 +139,17 @@ svuint16_t test_svget2_u16(svuint16x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_u3212svuint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP2]]
 //
 svuint32_t test_svget2_u32(svuint32x2_t tuple) ATTR
 {
@@ -130,13 +158,17 @@ svuint32_t test_svget2_u32(svuint32x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[TUPLE:%.*]], i64 2)
-// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_u6412svuint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[TUPLE:%.*]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP2]]
 //
 svuint64_t test_svget2_u64(svuint64x2_t tuple) ATTR
 {
@@ -145,13 +177,17 @@ svuint64_t test_svget2_u64(svuint64x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_f1613svfloat16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
 //
 svfloat16_t test_svget2_f16(svfloat16x2_t tuple) ATTR
 {
@@ -160,13 +196,17 @@ svfloat16_t test_svget2_f16(svfloat16x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[TUPLE:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_f3213svfloat32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[TUPLE:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP2]]
 //
 svfloat32_t test_svget2_f32(svfloat32x2_t tuple) ATTR
 {
@@ -175,13 +215,17 @@ svfloat32_t test_svget2_f32(svfloat32x2_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget2_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget2_f6413svfloat64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP2]]
 //
 svfloat64_t test_svget2_f64(svfloat64x2_t tuple) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3-bfloat.c
index c89f686c6bd2ae..950c4dad9749af 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3-bfloat.c
@@ -24,13 +24,19 @@
 
 // CHECK-LABEL: @test_svget3_bf16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv24bf16(<vscale x 24 x bfloat> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget3_bf16_014svbfloat16x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv24bf16(<vscale x 24 x bfloat> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP3]]
 //
 svbfloat16_t test_svget3_bf16_0(svbfloat16x3_t tuple) ATTR
 {
@@ -39,13 +45,19 @@ svbfloat16_t test_svget3_bf16_0(svbfloat16x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_bf16_1(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv24bf16(<vscale x 24 x bfloat> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget3_bf16_114svbfloat16x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv24bf16(<vscale x 24 x bfloat> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP3]]
 //
 svbfloat16_t test_svget3_bf16_1(svbfloat16x3_t tuple) ATTR
 {
@@ -54,13 +66,19 @@ svbfloat16_t test_svget3_bf16_1(svbfloat16x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_bf16_2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv24bf16(<vscale x 24 x bfloat> [[TUPLE:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 2
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget3_bf16_214svbfloat16x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv24bf16(<vscale x 24 x bfloat> [[TUPLE:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP3]]
 //
 svbfloat16_t test_svget3_bf16_2(svbfloat16x3_t tuple) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3.c
index a7f74d29ed2904..d85121262f28e2 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get3.c
@@ -26,13 +26,19 @@
 
 // CHECK-LABEL: @test_svget3_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv48i8(<vscale x 48 x i8> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svget3_s810svint8x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv48i8(<vscale x 48 x i8> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 svint8_t test_svget3_s8(svint8x3_t tuple) ATTR
 {
@@ -41,13 +47,19 @@ svint8_t test_svget3_s8(svint8x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv24i16(<vscale x 24 x i16> [[TUPLE:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_s1611svint16x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv24i16(<vscale x 24 x i16> [[TUPLE:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP3]]
 //
 svint16_t test_svget3_s16(svint16x3_t tuple) ATTR
 {
@@ -56,13 +68,19 @@ svint16_t test_svget3_s16(svint16x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv12i32(<vscale x 12 x i32> [[TUPLE:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_s3211svint32x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv12i32(<vscale x 12 x i32> [[TUPLE:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 //
 svint32_t test_svget3_s32(svint32x3_t tuple) ATTR
 {
@@ -71,13 +89,19 @@ svint32_t test_svget3_s32(svint32x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv6i64(<vscale x 6 x i64> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
+// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_s6411svint64x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv6i64(<vscale x 6 x i64> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 //
 svint64_t test_svget3_s64(svint64x3_t tuple) ATTR
 {
@@ -86,13 +110,19 @@ svint64_t test_svget3_s64(svint64x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv48i8(<vscale x 48 x i8> [[TUPLE:%.*]], i64 32)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svget3_u811svuint8x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv48i8(<vscale x 48 x i8> [[TUPLE:%.*]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP3]]
 //
 svuint8_t test_svget3_u8(svuint8x3_t tuple) ATTR
 {
@@ -101,13 +131,19 @@ svuint8_t test_svget3_u8(svuint8x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv24i16(<vscale x 24 x i16> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_u1612svuint16x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv24i16(<vscale x 24 x i16> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP3]]
 //
 svuint16_t test_svget3_u16(svuint16x3_t tuple) ATTR
 {
@@ -116,13 +152,19 @@ svuint16_t test_svget3_u16(svuint16x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv12i32(<vscale x 12 x i32> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_u3212svuint32x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv12i32(<vscale x 12 x i32> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP3]]
 //
 svuint32_t test_svget3_u32(svuint32x3_t tuple) ATTR
 {
@@ -131,13 +173,19 @@ svuint32_t test_svget3_u32(svuint32x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv6i64(<vscale x 6 x i64> [[TUPLE:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
+// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_u6412svuint64x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv6i64(<vscale x 6 x i64> [[TUPLE:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP3]]
 //
 svuint64_t test_svget3_u64(svuint64x3_t tuple) ATTR
 {
@@ -146,13 +194,19 @@ svuint64_t test_svget3_u64(svuint64x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv24f16(<vscale x 24 x half> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_f1613svfloat16x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv24f16(<vscale x 24 x half> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP3]]
 //
 svfloat16_t test_svget3_f16(svfloat16x3_t tuple) ATTR
 {
@@ -161,13 +215,19 @@ svfloat16_t test_svget3_f16(svfloat16x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv12f32(<vscale x 12 x float> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
+// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_f3213svfloat32x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv12f32(<vscale x 12 x float> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP3]]
 //
 svfloat32_t test_svget3_f32(svfloat32x3_t tuple) ATTR
 {
@@ -176,13 +236,19 @@ svfloat32_t test_svget3_f32(svfloat32x3_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget3_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv6f64(<vscale x 6 x double> [[TUPLE:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 2
+// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP3]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget3_f6413svfloat64x3_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv6f64(<vscale x 6 x double> [[TUPLE:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP3]]
 //
 svfloat64_t test_svget3_f64(svfloat64x3_t tuple) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4-bfloat.c
index 79eed6912e3ced..55b379140ca3dd 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4-bfloat.c
@@ -24,13 +24,21 @@
 
 // CHECK-LABEL: @test_svget4_bf16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget4_bf16_014svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 svbfloat16_t test_svget4_bf16_0(svbfloat16x4_t tuple) ATTR
 {
@@ -39,13 +47,21 @@ svbfloat16_t test_svget4_bf16_0(svbfloat16x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_bf16_1(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget4_bf16_114svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 svbfloat16_t test_svget4_bf16_1(svbfloat16x4_t tuple) ATTR
 {
@@ -54,13 +70,21 @@ svbfloat16_t test_svget4_bf16_1(svbfloat16x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_bf16_2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 2
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget4_bf16_214svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 svbfloat16_t test_svget4_bf16_2(svbfloat16x4_t tuple) ATTR
 {
@@ -69,13 +93,21 @@ svbfloat16_t test_svget4_bf16_2(svbfloat16x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_bf16_3(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 24)
-// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 3
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svget4_bf16_314svbfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[TUPLE:%.*]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } poison, <vscale x 8 x bfloat> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP0]], <vscale x 8 x bfloat> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], <vscale x 8 x bfloat> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], <vscale x 8 x bfloat> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP4]]
 //
 svbfloat16_t test_svget4_bf16_3(svbfloat16x4_t tuple) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4.c
index 278c217a0def41..d3eaae22a373fa 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_get4.c
@@ -27,13 +27,21 @@
 // several parameters, one for each member of the original struct.
 // CHECK-LABEL: @test_svget4_s8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]], 0
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svget4_s810svint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
 //
 svint8_t test_svget4_s8(svint8x4_t tuple) ATTR
 {
@@ -42,13 +50,21 @@ svint8_t test_svget4_s8(svint8x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[TUPLE:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_s1611svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[TUPLE:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 svint16_t test_svget4_s16(svint16x4_t tuple) ATTR
 {
@@ -57,13 +73,21 @@ svint16_t test_svget4_s16(svint16x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[TUPLE:%.*]], i64 8)
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_s3211svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[TUPLE:%.*]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 svint32_t test_svget4_s32(svint32x4_t tuple) ATTR
 {
@@ -72,13 +96,21 @@ svint32_t test_svget4_s32(svint32x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_s64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[TUPLE:%.*]], i64 6)
-// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_s6411svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[TUPLE:%.*]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 svint64_t test_svget4_s64(svint64x4_t tuple) ATTR
 {
@@ -87,13 +119,21 @@ svint64_t test_svget4_s64(svint64x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_u8(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[TUPLE:%.*]], i64 32)
-// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]], 2
+// CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svget4_u811svuint8x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.vector.extract.nxv16i8.nxv64i8(<vscale x 64 x i8> [[TUPLE:%.*]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } poison, <vscale x 16 x i8> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], <vscale x 16 x i8> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], <vscale x 16 x i8> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP2]], <vscale x 16 x i8> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP4]]
 //
 svuint8_t test_svget4_u8(svuint8x4_t tuple) ATTR
 {
@@ -102,13 +142,21 @@ svuint8_t test_svget4_u8(svuint8x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[TUPLE:%.*]], i64 24)
-// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 3
+// CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_u1612svuint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[TUPLE:%.*]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], <vscale x 8 x i16> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], <vscale x 8 x i16> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP4]]
 //
 svuint16_t test_svget4_u16(svuint16x4_t tuple) ATTR
 {
@@ -117,13 +165,21 @@ svuint16_t test_svget4_u16(svuint16x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_u3212svuint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], <vscale x 4 x i32> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], <vscale x 4 x i32> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP4]]
 //
 svuint32_t test_svget4_u32(svuint32x4_t tuple) ATTR
 {
@@ -132,13 +188,21 @@ svuint32_t test_svget4_u32(svuint32x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_u64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[TUPLE:%.*]], i64 6)
-// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_u6412svuint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[TUPLE:%.*]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], <vscale x 2 x i64> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], <vscale x 2 x i64> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP4]]
 //
 svuint64_t test_svget4_u64(svuint64x4_t tuple) ATTR
 {
@@ -147,13 +211,21 @@ svuint64_t test_svget4_u64(svuint64x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_f16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[TUPLE:%.*]], i64 16)
-// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_f1613svfloat16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[TUPLE:%.*]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 8 x half> [[TMP4]]
 //
 svfloat16_t test_svget4_f16(svfloat16x4_t tuple) ATTR
 {
@@ -162,13 +234,21 @@ svfloat16_t test_svget4_f16(svfloat16x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[TUPLE:%.*]], i64 0)
-// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_f3213svfloat32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[TUPLE:%.*]], i64 0)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP4]]
 //
 svfloat32_t test_svget4_f32(svfloat32x4_t tuple) ATTR
 {
@@ -177,13 +257,21 @@ svfloat32_t test_svget4_f32(svfloat32x4_t tuple) ATTR
 
 // CHECK-LABEL: @test_svget4_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[TUPLE:%.*]], i64 4)
-// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP0]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[TUPLE_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[TUPLE_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[TUPLE_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[TUPLE_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CHECK-NEXT:    ret <vscale x 2 x double> [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svget4_f6413svfloat64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[TUPLE:%.*]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP0]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[TUPLE_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[TUPLE_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[TUPLE_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[TUPLE_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    ret <vscale x 2 x double> [[TMP4]]
 //
 svfloat64_t test_svget4_f64(svfloat64x4_t tuple) ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c
index fcae89c50b0e65..5535b3d090d323 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2-bfloat.c
@@ -26,21 +26,13 @@
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svld2_bf16u10__SVBool_tPKu6__bf16(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 svbfloat16x2_t test_svld2_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
 {
@@ -53,22 +45,14 @@ svbfloat16x2_t test_svld2_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svld2_vnum_bf16u10__SVBool_tPKu6__bf16l(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 svbfloat16x2_t test_svld2_vnum_bf16(svbool_t pg, const bfloat16_t *base, int64_t vnum) MODE_ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c
index 992b51fa251233..3097cb9cbcaab4 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld2.c
@@ -26,20 +26,12 @@
 // CHECK-LABEL: @test_svld2_s8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld2_s8u10__SVBool_tPKa(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 svint8x2_t test_svld2_s8(svbool_t pg, const int8_t *base) MODE_ATTR
 {
@@ -50,21 +42,13 @@ svint8x2_t test_svld2_s8(svbool_t pg, const int8_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_s16u10__SVBool_tPKs(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 svint16x2_t test_svld2_s16(svbool_t pg, const int16_t *base) MODE_ATTR
 {
@@ -75,21 +59,13 @@ svint16x2_t test_svld2_s16(svbool_t pg, const int16_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_s32u10__SVBool_tPKi(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 svint32x2_t test_svld2_s32(svbool_t pg, const int32_t *base) MODE_ATTR
 {
@@ -100,21 +76,13 @@ svint32x2_t test_svld2_s32(svbool_t pg, const int32_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_s64u10__SVBool_tPKl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 svint64x2_t test_svld2_s64(svbool_t pg, const int64_t *base) MODE_ATTR
 {
@@ -124,20 +92,12 @@ svint64x2_t test_svld2_s64(svbool_t pg, const int64_t *base) MODE_ATTR
 // CHECK-LABEL: @test_svld2_u8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld2_u8u10__SVBool_tPKh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP4]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 svuint8x2_t test_svld2_u8(svbool_t pg, const uint8_t *base) MODE_ATTR
 {
@@ -148,21 +108,13 @@ svuint8x2_t test_svld2_u8(svbool_t pg, const uint8_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_u16u10__SVBool_tPKt(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 svuint16x2_t test_svld2_u16(svbool_t pg, const uint16_t *base) MODE_ATTR
 {
@@ -173,21 +125,13 @@ svuint16x2_t test_svld2_u16(svbool_t pg, const uint16_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_u32u10__SVBool_tPKj(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 svuint32x2_t test_svld2_u32(svbool_t pg, const uint32_t *base) MODE_ATTR
 {
@@ -198,21 +142,13 @@ svuint32x2_t test_svld2_u32(svbool_t pg, const uint32_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_u64u10__SVBool_tPKm(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 svuint64x2_t test_svld2_u64(svbool_t pg, const uint64_t *base) MODE_ATTR
 {
@@ -223,21 +159,13 @@ svuint64x2_t test_svld2_u64(svbool_t pg, const uint64_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_f16u10__SVBool_tPKDh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]]
 //
 svfloat16x2_t test_svld2_f16(svbool_t pg, const float16_t *base) MODE_ATTR
 {
@@ -248,21 +176,13 @@ svfloat16x2_t test_svld2_f16(svbool_t pg, const float16_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_f32u10__SVBool_tPKf(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]]
 //
 svfloat32x2_t test_svld2_f32(svbool_t pg, const float32_t *base) MODE_ATTR
 {
@@ -273,21 +193,13 @@ svfloat32x2_t test_svld2_f32(svbool_t pg, const float32_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld2_f64u10__SVBool_tPKd(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]]
 //
 svfloat64x2_t test_svld2_f64(svbool_t pg, const float64_t *base) MODE_ATTR
 {
@@ -298,21 +210,13 @@ svfloat64x2_t test_svld2_f64(svbool_t pg, const float64_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld2_vnum_s8u10__SVBool_tPKal(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 svint8x2_t test_svld2_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum) MODE_ATTR
 {
@@ -324,22 +228,14 @@ svint8x2_t test_svld2_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum) MOD
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_s16u10__SVBool_tPKsl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 svint16x2_t test_svld2_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum) MODE_ATTR
 {
@@ -351,22 +247,14 @@ svint16x2_t test_svld2_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum)
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_s32u10__SVBool_tPKil(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 svint32x2_t test_svld2_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum) MODE_ATTR
 {
@@ -378,22 +266,14 @@ svint32x2_t test_svld2_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum)
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_s64u10__SVBool_tPKll(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 svint64x2_t test_svld2_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum) MODE_ATTR
 {
@@ -404,21 +284,13 @@ svint64x2_t test_svld2_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum)
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP5]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld2_vnum_u8u10__SVBool_tPKhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.insert.nxv32i8.nxv16i8(<vscale x 32 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x i8> [[TMP5]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 svuint8x2_t test_svld2_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum) MODE_ATTR
 {
@@ -430,22 +302,14 @@ svuint8x2_t test_svld2_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum) M
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_u16u10__SVBool_tPKtl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x i16> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 svuint16x2_t test_svld2_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum) MODE_ATTR
 {
@@ -457,22 +321,14 @@ svuint16x2_t test_svld2_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_u32u10__SVBool_tPKjl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x i32> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 svuint32x2_t test_svld2_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum) MODE_ATTR
 {
@@ -484,22 +340,14 @@ svuint32x2_t test_svld2_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_u64u10__SVBool_tPKml(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x i64> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 svuint64x2_t test_svld2_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum) MODE_ATTR
 {
@@ -511,22 +359,14 @@ svuint64x2_t test_svld2_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x half>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_f16u10__SVBool_tPKDhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x half>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]]
 //
 svfloat16x2_t test_svld2_vnum_f16(svbool_t pg, const float16_t *base, int64_t vnum) MODE_ATTR
 {
@@ -538,22 +378,14 @@ svfloat16x2_t test_svld2_vnum_f16(svbool_t pg, const float16_t *base, int64_t vn
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x float>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_f32u10__SVBool_tPKfl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x float>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]]
 //
 svfloat32x2_t test_svld2_vnum_f32(svbool_t pg, const float32_t *base, int64_t vnum) MODE_ATTR
 {
@@ -565,22 +397,14 @@ svfloat32x2_t test_svld2_vnum_f32(svbool_t pg, const float32_t *base, int64_t vn
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x double>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld2_vnum_f64u10__SVBool_tPKdl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x double>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]]
 //
 svfloat64x2_t test_svld2_vnum_f64(svbool_t pg, const float64_t *base, int64_t vnum) MODE_ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c
index 9100d27534c1ca..41a367b737b0f7 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3-bfloat.c
@@ -27,25 +27,13 @@
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svld3_bf16u10__SVBool_tPKu6__bf16(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 svbfloat16x3_t test_svld3_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
 {
@@ -57,26 +45,14 @@ svbfloat16x3_t test_svld3_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svld3_vnum_bf16u10__SVBool_tPKu6__bf16l(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x bfloat> @llvm.vector.insert.nxv24bf16.nxv8bf16(<vscale x 24 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x bfloat> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 svbfloat16x3_t test_svld3_vnum_bf16(svbool_t pg, const bfloat16_t *base, int64_t vnum) MODE_ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c
index 10206b5362e11e..2deb5a1d4930cf 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld3.c
@@ -25,24 +25,12 @@
 // CHECK-LABEL: @test_svld3_s8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld3_s8u10__SVBool_tPKa(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 svint8x3_t test_svld3_s8(svbool_t pg, const int8_t *base) MODE_ATTR
 {
@@ -53,25 +41,13 @@ svint8x3_t test_svld3_s8(svbool_t pg, const int8_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_s16u10__SVBool_tPKs(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 svint16x3_t test_svld3_s16(svbool_t pg, const int16_t *base) MODE_ATTR
 {
@@ -82,25 +58,13 @@ svint16x3_t test_svld3_s16(svbool_t pg, const int16_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_s32u10__SVBool_tPKi(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 svint32x3_t test_svld3_s32(svbool_t pg, const int32_t *base) MODE_ATTR
 {
@@ -111,25 +75,13 @@ svint32x3_t test_svld3_s32(svbool_t pg, const int32_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_s64u10__SVBool_tPKl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 svint64x3_t test_svld3_s64(svbool_t pg, const int64_t *base) MODE_ATTR
 {
@@ -139,24 +91,12 @@ svint64x3_t test_svld3_s64(svbool_t pg, const int64_t *base) MODE_ATTR
 // CHECK-LABEL: @test_svld3_u8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld3_u8u10__SVBool_tPKh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP6]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 svuint8x3_t test_svld3_u8(svbool_t pg, const uint8_t *base) MODE_ATTR
 {
@@ -167,25 +107,13 @@ svuint8x3_t test_svld3_u8(svbool_t pg, const uint8_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_u16u10__SVBool_tPKt(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 svuint16x3_t test_svld3_u16(svbool_t pg, const uint16_t *base) MODE_ATTR
 {
@@ -196,25 +124,13 @@ svuint16x3_t test_svld3_u16(svbool_t pg, const uint16_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_u32u10__SVBool_tPKj(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP4]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP1]]
 //
 svuint32x3_t test_svld3_u32(svbool_t pg, const uint32_t *base) MODE_ATTR
 {
@@ -225,25 +141,13 @@ svuint32x3_t test_svld3_u32(svbool_t pg, const uint32_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_u64u10__SVBool_tPKm(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP1]]
 //
 svuint64x3_t test_svld3_u64(svbool_t pg, const uint64_t *base) MODE_ATTR
 {
@@ -254,25 +158,13 @@ svuint64x3_t test_svld3_u64(svbool_t pg, const uint64_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x half> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_f16u10__SVBool_tPKDh(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x half> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]]
 //
 svfloat16x3_t test_svld3_f16(svbool_t pg, const float16_t *base) MODE_ATTR
 {
@@ -283,25 +175,13 @@ svfloat16x3_t test_svld3_f16(svbool_t pg, const float16_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x float> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_f32u10__SVBool_tPKf(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x float> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]]
 //
 svfloat32x3_t test_svld3_f32(svbool_t pg, const float32_t *base) MODE_ATTR
 {
@@ -312,25 +192,13 @@ svfloat32x3_t test_svld3_f32(svbool_t pg, const float32_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x double> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld3_f64u10__SVBool_tPKd(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x double> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]]
 //
 svfloat64x3_t test_svld3_f64(svbool_t pg, const float64_t *base) MODE_ATTR
 {
@@ -341,25 +209,13 @@ svfloat64x3_t test_svld3_f64(svbool_t pg, const float64_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], i64 32)
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld3_vnum_s8u10__SVBool_tPKal(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 svint8x3_t test_svld3_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum) MODE_ATTR
 {
@@ -371,26 +227,14 @@ svint8x3_t test_svld3_vnum_s8(svbool_t pg, const int8_t *base, int64_t vnum) MOD
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_s16u10__SVBool_tPKsl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 svint16x3_t test_svld3_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum) MODE_ATTR
 {
@@ -402,26 +246,14 @@ svint16x3_t test_svld3_vnum_s16(svbool_t pg, const int16_t *base, int64_t vnum)
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_s32u10__SVBool_tPKil(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 svint32x3_t test_svld3_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum) MODE_ATTR
 {
@@ -433,26 +265,14 @@ svint32x3_t test_svld3_vnum_s32(svbool_t pg, const int32_t *base, int64_t vnum)
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_s64u10__SVBool_tPKll(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 svint64x3_t test_svld3_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum) MODE_ATTR
 {
@@ -463,25 +283,13 @@ svint64x3_t test_svld3_vnum_s64(svbool_t pg, const int64_t *base, int64_t vnum)
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], i64 32)
-// CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP7]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z18test_svld3_vnum_u8u10__SVBool_tPKhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = getelementptr <vscale x 16 x i8>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[TMP0]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> poison, <vscale x 16 x i8> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP3]], <vscale x 16 x i8> [[TMP4]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 48 x i8> @llvm.vector.insert.nxv48i8.nxv16i8(<vscale x 48 x i8> [[TMP5]], <vscale x 16 x i8> [[TMP6]], i64 32)
-// CPP-CHECK-NEXT:    ret <vscale x 48 x i8> [[TMP7]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP1]]
 //
 svuint8x3_t test_svld3_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum) MODE_ATTR
 {
@@ -493,26 +301,14 @@ svuint8x3_t test_svld3_vnum_u8(svbool_t pg, const uint8_t *base, int64_t vnum) M
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_u16u10__SVBool_tPKtl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x i16>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> poison, <vscale x 8 x i16> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x i16> @llvm.vector.insert.nxv24i16.nxv8i16(<vscale x 24 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x i16> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]]
 //
 svuint16x3_t test_svld3_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum) MODE_ATTR
 {
@@ -524,26 +320,14 @@ svuint16x3_t test_svld3_vnum_u16(svbool_t pg, const uint16_t *base, int64_t vnum
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_u32u10__SVBool_tPKjl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 12 x i32> @llvm.vector.insert.nxv12i32.nxv4i32(<vscale x 12 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x i32> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]]
 //
 svuint32x3_t test_svld3_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum) MODE_ATTR
 {
@@ -555,26 +339,14 @@ svuint32x3_t test_svld3_vnum_u32(svbool_t pg, const uint32_t *base, int64_t vnum
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_u64u10__SVBool_tPKml(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> poison, <vscale x 2 x i64> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 6 x i64> @llvm.vector.insert.nxv6i64.nxv2i64(<vscale x 6 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x i64> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]]
 //
 svuint64x3_t test_svld3_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum) MODE_ATTR
 {
@@ -586,26 +358,14 @@ svuint64x3_t test_svld3_vnum_u64(svbool_t pg, const uint64_t *base, int64_t vnum
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x half>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 16)
-// CHECK-NEXT:    ret <vscale x 24 x half> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_f16u10__SVBool_tPKDhl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x half>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 24 x half> @llvm.vector.insert.nxv24f16.nxv8f16(<vscale x 24 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    ret <vscale x 24 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]]
 //
 svfloat16x3_t test_svld3_vnum_f16(svbool_t pg, const float16_t *base, int64_t vnum) MODE_ATTR
 {
@@ -617,26 +377,14 @@ svfloat16x3_t test_svld3_vnum_f16(svbool_t pg, const float16_t *base, int64_t vn
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x float>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 12 x float> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_f32u10__SVBool_tPKfl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 4 x float>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 12 x float> @llvm.vector.insert.nxv12f32.nxv4f32(<vscale x 12 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 12 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]]
 //
 svfloat32x3_t test_svld3_vnum_f32(svbool_t pg, const float32_t *base, int64_t vnum) MODE_ATTR
 {
@@ -648,26 +396,14 @@ svfloat32x3_t test_svld3_vnum_f32(svbool_t pg, const float32_t *base, int64_t vn
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x double>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 6 x double> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svld3_vnum_f64u10__SVBool_tPKdl(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 2 x double>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 6 x double> @llvm.vector.insert.nxv6f64.nxv2f64(<vscale x 6 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 6 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]]
 //
 svfloat64x3_t test_svld3_vnum_f64(svbool_t pg, const float64_t *base, int64_t vnum) MODE_ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c
index 0f21ffdb6f7093..a88e6f11a05108 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4-bfloat.c
@@ -26,29 +26,13 @@
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
-// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 3
-// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP9]]
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z15test_svld4_bf16u10__SVBool_tPKu6__bf16(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 2
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]], 3
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP7]], <vscale x 8 x bfloat> [[TMP8]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP9]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP1]]
 //
 svbfloat16x4_t test_svld4_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
 {
@@ -60,30 +44,14 @@ svbfloat16x4_t test_svld4_bf16(svbool_t pg, const bfloat16_t *base) MODE_ATTR
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 2
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 16)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 3
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP10]]
+// CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 // CPP-CHECK-LABEL: @_Z20test_svld4_vnum_bf16u10__SVBool_tPKu6__bf16l(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = getelementptr <vscale x 8 x bfloat>, ptr [[BASE:%.*]], i64 [[VNUM:%.*]]
 // CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> [[TMP0]], ptr [[TMP1]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 2
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 3
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP10]]
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]]
 //
 svbfloat16x4_t test_svld4_vnum_bf16(svbool_t pg, const bfloat16_t *base, int64_t vnum) MODE_ATTR
 {

diff  --git a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c
index 06e07307886534..30796a4f46a724 100644
--- a/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c
+++ b/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld4.c
@@ -25,28 +25,12 @@
 // CHECK-LABEL: @test_svld4_s8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z13test_svld4_s8u10__SVBool_tPKa(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> [[PG:%.*]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 0
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> poison, <vscale x 16 x i8> [[TMP1]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 1
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP2]], <vscale x 16 x i8> [[TMP3]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 2
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP4]], <vscale x 16 x i8> [[TMP5]], i64 32)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]], 3
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 64 x i8> @llvm.vector.insert.nxv64i8.nxv16i8(<vscale x 64 x i8> [[TMP6]], <vscale x 16 x i8> [[TMP7]], i64 48)
-// CPP-CHECK-NEXT:    ret <vscale x 64 x i8> [[TMP8]]
+// CPP-CHECK-NEXT:    ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } [[TMP0]]
 //
 svint8x4_t test_svld4_s8(svbool_t pg, const int8_t *base) MODE_ATTR
 {
@@ -57,29 +41,13 @@ svint8x4_t test_svld4_s8(svbool_t pg, const int8_t *base) MODE_ATTR
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 2
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], i64 16)
-// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 3
-// CHECK-NEXT:    [[TMP9:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP7]], <vscale x 8 x i16> [[TMP8]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x i16> [[TMP9]]
+// CHECK-NEXT:    ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]]
 //
 // CPP-CHECK-LABEL: @_Z14test_svld4_s16u10__SVBool_tPKs(
 // CPP-CHECK-NEXT:  entry:
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
 // CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> [[TMP0]], ptr [[BASE:%.*]])
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 0
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> poison, <vscale x 8 x i16> [[TMP2]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP1]], 1
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 32 x i16> @llvm.vector.insert.nxv32i16.nxv8i16(<vscale x 32 x i16> [[TMP3]], <vscale x 8 x i16> [[TMP4]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] =