[llvm] [AArch64][SVE] Instcombine uzp1/reinterpret svbool to use vector.insert (PR #81069)

via llvm-commits llvm-commits at lists.llvm.org
Wed Feb 7 17:08:32 PST 2024


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Usman Nadeem (UsmanNadeem)

<details>
<summary>Changes</summary>

Concatenating two predictes using uzp1 after converting to double length
using sve.convert.to/from.svbool is optimized poorly in the backend,
resulting in additional `and` instructions to zero the lanes. See
https://github.com/llvm/llvm-project/pull/78623/

Combine this pattern to use `llvm.vector.insert` to concatenate and get
rid of convert to/from svbools.

Change-Id: Ieb38055f1f2304ead1ab2bead03ebd046b3bd36f


---
Full diff: https://github.com/llvm/llvm-project/pull/81069.diff


2 Files Affected:

- (modified) llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp (+32) 
- (added) llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-opts-uzp1.ll (+177) 


``````````diff
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index cdd2750521d2c9..d583b38d9ff750 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -1630,6 +1630,36 @@ static std::optional<Instruction *> instCombineSVETBL(InstCombiner &IC,
   return IC.replaceInstUsesWith(II, VectorSplat);
 }
 
+static std::optional<Instruction *> instCombineSVEUzp1(InstCombiner &IC,
+                                                       IntrinsicInst &II) {
+  Value *A, *B;
+  Type *RetTy = II.getType();
+  constexpr Intrinsic::ID From_SV = Intrinsic::aarch64_sve_convert_from_svbool;
+  constexpr Intrinsic::ID To_SV = Intrinsic::aarch64_sve_convert_to_svbool;
+
+  // uzp1(to_svbool(A), to_svbool(B)) --> <A, B>
+  // uzp1(from_svbool(to_svbool(A)), from_svbool(to_svbool(B))) --> <A, B>
+  if ((match(II.getArgOperand(0),
+             m_Intrinsic<From_SV>(m_Intrinsic<To_SV>(m_Value(A)))) &&
+       match(II.getArgOperand(1),
+             m_Intrinsic<From_SV>(m_Intrinsic<To_SV>(m_Value(B))))) ||
+      (match(II.getArgOperand(0), m_Intrinsic<To_SV>(m_Value(A))) &&
+       match(II.getArgOperand(1), m_Intrinsic<To_SV>(m_Value(B))))) {
+    auto *TyA = cast<ScalableVectorType>(A->getType());
+    if (TyA == B->getType() &&
+        RetTy == ScalableVectorType::getDoubleElementsVectorType(TyA)) {
+      auto *SubVec = IC.Builder.CreateInsertVector(
+          RetTy, PoisonValue::get(RetTy), A, IC.Builder.getInt64(0));
+      auto *ConcatVec = IC.Builder.CreateInsertVector(
+          RetTy, SubVec, B, IC.Builder.getInt64(TyA->getMinNumElements()));
+      ConcatVec->takeName(&II);
+      return IC.replaceInstUsesWith(II, ConcatVec);
+    }
+  }
+
+  return std::nullopt;
+}
+
 static std::optional<Instruction *> instCombineSVEZip(InstCombiner &IC,
                                                       IntrinsicInst &II) {
   // zip1(uzp1(A, B), uzp2(A, B)) --> A
@@ -2012,6 +2042,8 @@ AArch64TTIImpl::instCombineIntrinsic(InstCombiner &IC,
   case Intrinsic::aarch64_sve_sunpkhi:
   case Intrinsic::aarch64_sve_sunpklo:
     return instCombineSVEUnpack(IC, II);
+  case Intrinsic::aarch64_sve_uzp1:
+    return instCombineSVEUzp1(IC, II);
   case Intrinsic::aarch64_sve_zip1:
   case Intrinsic::aarch64_sve_zip2:
     return instCombineSVEZip(IC, II);
diff --git a/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-opts-uzp1.ll b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-opts-uzp1.ll
new file mode 100644
index 00000000000000..7fc59af6edd458
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-opts-uzp1.ll
@@ -0,0 +1,177 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -S -passes=instcombine -mtriple=aarch64 < %s | FileCheck %s
+
+; Code the concatenates two predictes using uzp1 after converting to
+; double length using sve.convert.to/from.svbool is optimized poorly
+; in the backend, resulting in additional `and` instructions to zero
+; the lanes. Test that we get rid of convert to/from and generate a
+; concatenate using vector insert instead.
+
+
+define <vscale x 8 x i1> @reinterpt_uzp1_1(<vscale x 4 x i32> %v0, <vscale x 4 x i32> %v1, <vscale x 4 x i32> %x) {
+; CHECK-LABEL: define <vscale x 8 x i1> @reinterpt_uzp1_1(
+; CHECK-SAME: <vscale x 4 x i32> [[V0:%.*]], <vscale x 4 x i32> [[V1:%.*]], <vscale x 4 x i32> [[X:%.*]]) {
+; CHECK-NEXT:    [[CMP0:%.*]] = icmp ult <vscale x 4 x i32> [[V0]], [[X]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult <vscale x 4 x i32> [[V1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> poison, <vscale x 4 x i1> [[CMP0]], i64 0)
+; CHECK-NEXT:    [[UZ1:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> [[TMP1]], <vscale x 4 x i1> [[CMP1]], i64 4)
+; CHECK-NEXT:    ret <vscale x 8 x i1> [[UZ1]]
+;
+  %cmp0 = icmp ult <vscale x 4 x i32> %v0, %x
+  %cmp1 = icmp ult <vscale x 4 x i32> %v1, %x
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp0)
+  %2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %1)
+  %3 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp1)
+  %4 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %3)
+  %uz1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> %2, <vscale x 8 x i1> %4)
+  ret <vscale x 8 x i1> %uz1
+}
+
+define <vscale x 8 x i1> @reinterpt_uzp1_2(<vscale x 2 x i64> %v0, <vscale x 2 x i64> %v1, <vscale x 2 x i64> %x) {
+; CHECK-LABEL: define <vscale x 8 x i1> @reinterpt_uzp1_2(
+; CHECK-SAME: <vscale x 2 x i64> [[V0:%.*]], <vscale x 2 x i64> [[V1:%.*]], <vscale x 2 x i64> [[X:%.*]]) {
+; CHECK-NEXT:    [[CMP0:%.*]] = icmp ult <vscale x 2 x i64> [[V0]], [[X]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult <vscale x 2 x i64> [[V1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[CMP0]])
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[CMP1]])
+; CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[TMP3]])
+; CHECK-NEXT:    [[TMP5:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> poison, <vscale x 4 x i1> [[TMP2]], i64 0)
+; CHECK-NEXT:    [[UZ1:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> [[TMP5]], <vscale x 4 x i1> [[TMP4]], i64 4)
+; CHECK-NEXT:    ret <vscale x 8 x i1> [[UZ1]]
+;
+  %cmp0 = icmp ult <vscale x 2 x i64> %v0, %x
+  %cmp1 = icmp ult <vscale x 2 x i64> %v1, %x
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %cmp0)
+  %2 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %1)
+  %3 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %cmp1)
+  %4 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %3)
+  %5 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %2)
+  %6 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %5)
+  %7 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %4)
+  %8 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %7)
+  %uz1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> %6, <vscale x 8 x i1> %8)
+  ret <vscale x 8 x i1> %uz1
+}
+
+define <vscale x 16 x i1> @reinterpt_uzp1_3(<vscale x 4 x i32> %v0, <vscale x 4 x i32> %v1, <vscale x 4 x i32> %v2, <vscale x 4 x i32> %v3, <vscale x 4 x i32> %x) {
+; CHECK-LABEL: define <vscale x 16 x i1> @reinterpt_uzp1_3(
+; CHECK-SAME: <vscale x 4 x i32> [[V0:%.*]], <vscale x 4 x i32> [[V1:%.*]], <vscale x 4 x i32> [[V2:%.*]], <vscale x 4 x i32> [[V3:%.*]], <vscale x 4 x i32> [[X:%.*]]) {
+; CHECK-NEXT:    [[CMP0:%.*]] = icmp ult <vscale x 4 x i32> [[V0]], [[X]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult <vscale x 4 x i32> [[V1]], [[X]]
+; CHECK-NEXT:    [[CMP2:%.*]] = icmp ult <vscale x 4 x i32> [[V2]], [[X]]
+; CHECK-NEXT:    [[CMP3:%.*]] = icmp ult <vscale x 4 x i32> [[V3]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> poison, <vscale x 4 x i1> [[CMP0]], i64 0)
+; CHECK-NEXT:    [[UZ1_1:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> [[TMP1]], <vscale x 4 x i1> [[CMP1]], i64 4)
+; CHECK-NEXT:    [[TMP2:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> poison, <vscale x 4 x i1> [[CMP2]], i64 0)
+; CHECK-NEXT:    [[UZ1_2:%.*]] = call <vscale x 8 x i1> @llvm.vector.insert.nxv8i1.nxv4i1(<vscale x 8 x i1> [[TMP2]], <vscale x 4 x i1> [[CMP3]], i64 4)
+; CHECK-NEXT:    [[TMP3:%.*]] = call <vscale x 16 x i1> @llvm.vector.insert.nxv16i1.nxv8i1(<vscale x 16 x i1> poison, <vscale x 8 x i1> [[UZ1_1]], i64 0)
+; CHECK-NEXT:    [[UZ3:%.*]] = call <vscale x 16 x i1> @llvm.vector.insert.nxv16i1.nxv8i1(<vscale x 16 x i1> [[TMP3]], <vscale x 8 x i1> [[UZ1_2]], i64 8)
+; CHECK-NEXT:    ret <vscale x 16 x i1> [[UZ3]]
+;
+  %cmp0 = icmp ult <vscale x 4 x i32> %v0, %x
+  %cmp1 = icmp ult <vscale x 4 x i32> %v1, %x
+  %cmp2 = icmp ult <vscale x 4 x i32> %v2, %x
+  %cmp3 = icmp ult <vscale x 4 x i32> %v3, %x
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp0)
+  %2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %1)
+  %3 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp1)
+  %4 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %3)
+  %uz1_1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> %2, <vscale x 8 x i1> %4)
+  %5 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp2)
+  %6 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %5)
+  %7 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp3)
+  %8 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %7)
+  %uz1_2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> %6, <vscale x 8 x i1> %8)
+  %9 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %uz1_1)
+  %10 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %uz1_2)
+  %uz3 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.uzp1.nxv16i1(<vscale x 16 x i1> %9, <vscale x 16 x i1> %10)
+  ret <vscale x 16 x i1> %uz3
+}
+
+define <vscale x 4 x i1> @neg1(<vscale x 4 x i32> %v0, <vscale x 4 x i32> %v1, <vscale x 4 x i32> %x) {
+; CHECK-LABEL: define <vscale x 4 x i1> @neg1(
+; CHECK-SAME: <vscale x 4 x i32> [[V0:%.*]], <vscale x 4 x i32> [[V1:%.*]], <vscale x 4 x i32> [[X:%.*]]) {
+; CHECK-NEXT:    [[CMP0:%.*]] = icmp ult <vscale x 4 x i32> [[V0]], [[X]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult <vscale x 4 x i32> [[V1]], [[X]]
+; CHECK-NEXT:    [[UZ1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.uzp1.nxv4i1(<vscale x 4 x i1> [[CMP0]], <vscale x 4 x i1> [[CMP1]])
+; CHECK-NEXT:    ret <vscale x 4 x i1> [[UZ1]]
+;
+  %cmp0 = icmp ult <vscale x 4 x i32> %v0, %x
+  %cmp1 = icmp ult <vscale x 4 x i32> %v1, %x
+  %uz1 = tail call <vscale x 4 x i1> @llvm.aarch64.sve.uzp1.nxv4i1(<vscale x 4 x i1> %cmp0, <vscale x 4 x i1> %cmp1)
+  ret <vscale x 4 x i1> %uz1
+}
+
+define <vscale x 8 x i1> @neg2(<vscale x 2 x i64> %v0, <vscale x 4 x i32> %v1, <vscale x 4 x i32> %x, <vscale x 2 x i64> %y) {
+; CHECK-LABEL: define <vscale x 8 x i1> @neg2(
+; CHECK-SAME: <vscale x 2 x i64> [[V0:%.*]], <vscale x 4 x i32> [[V1:%.*]], <vscale x 4 x i32> [[X:%.*]], <vscale x 2 x i64> [[Y:%.*]]) {
+; CHECK-NEXT:    [[CMP0:%.*]] = icmp ult <vscale x 2 x i64> [[V0]], [[Y]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult <vscale x 4 x i32> [[V1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[CMP0]])
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[CMP1]])
+; CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[TMP3]])
+; CHECK-NEXT:    [[UZ1:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> [[TMP2]], <vscale x 8 x i1> [[TMP4]])
+; CHECK-NEXT:    ret <vscale x 8 x i1> [[UZ1]]
+;
+  %cmp0 = icmp ult <vscale x 2 x i64> %v0, %y
+  %cmp1 = icmp ult <vscale x 4 x i32> %v1, %x
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %cmp0)
+  %2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %1)
+  %3 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp1)
+  %4 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %3)
+  %uz1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> %2, <vscale x 8 x i1> %4)
+  ret <vscale x 8 x i1> %uz1
+}
+
+define <vscale x 8 x i1> @neg3(<vscale x 8 x i16> %v0, <vscale x 4 x i32> %v1, <vscale x 4 x i32> %x, <vscale x 8 x i16> %y) {
+; CHECK-LABEL: define <vscale x 8 x i1> @neg3(
+; CHECK-SAME: <vscale x 8 x i16> [[V0:%.*]], <vscale x 4 x i32> [[V1:%.*]], <vscale x 4 x i32> [[X:%.*]], <vscale x 8 x i16> [[Y:%.*]]) {
+; CHECK-NEXT:    [[CMP0:%.*]] = icmp ult <vscale x 8 x i16> [[V0]], [[Y]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult <vscale x 4 x i32> [[V1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[CMP1]])
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[TMP1]])
+; CHECK-NEXT:    [[UZ1:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> [[CMP0]], <vscale x 8 x i1> [[TMP2]])
+; CHECK-NEXT:    ret <vscale x 8 x i1> [[UZ1]]
+;
+  %cmp0 = icmp ult <vscale x 8 x i16> %v0, %y
+  %cmp1 = icmp ult <vscale x 4 x i32> %v1, %x
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %cmp1)
+  %2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %1)
+  %uz1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> %cmp0, <vscale x 8 x i1> %2)
+  ret <vscale x 8 x i1> %uz1
+}
+
+define <vscale x 8 x i1> @neg4(<vscale x 2 x i64> %v0, <vscale x 2 x i64> %v1, <vscale x 2 x i64> %x) {
+; CHECK-LABEL: define <vscale x 8 x i1> @neg4(
+; CHECK-SAME: <vscale x 2 x i64> [[V0:%.*]], <vscale x 2 x i64> [[V1:%.*]], <vscale x 2 x i64> [[X:%.*]]) {
+; CHECK-NEXT:    [[CMP0:%.*]] = icmp ult <vscale x 2 x i64> [[V0]], [[X]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp ult <vscale x 2 x i64> [[V1]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[CMP0]])
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[CMP1]])
+; CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[TMP3]])
+; CHECK-NEXT:    [[UZ1:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> [[TMP2]], <vscale x 8 x i1> [[TMP4]])
+; CHECK-NEXT:    ret <vscale x 8 x i1> [[UZ1]]
+;
+  %cmp0 = icmp ult <vscale x 2 x i64> %v0, %x
+  %cmp1 = icmp ult <vscale x 2 x i64> %v1, %x
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %cmp0)
+  %2 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %1)
+  %3 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %cmp1)
+  %4 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %3)
+  %uz1 = tail call <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1> %2, <vscale x 8 x i1> %4)
+  ret <vscale x 8 x i1> %uz1
+}
+
+declare <vscale x 4 x i1> @llvm.aarch64.sve.uzp1.nxv4i1(<vscale x 4 x i1>, <vscale x 4 x i1>)
+declare <vscale x 8 x i1> @llvm.aarch64.sve.uzp1.nxv8i1(<vscale x 8 x i1>, <vscale x 8 x i1>)
+declare <vscale x 16 x i1> @llvm.aarch64.sve.uzp1.nxv16i1(<vscale x 16 x i1>, <vscale x 16 x i1>)
+
+declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1>)
+declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1>)
+declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1>)
+declare <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1>)
+

``````````

</details>


https://github.com/llvm/llvm-project/pull/81069


More information about the llvm-commits mailing list