[clang] 9c95f37 - [CIR][AArch64] Lower NEON vuzp intrinsics (#195591)
via cfe-commits
cfe-commits at lists.llvm.org
Sun May 10 08:56:23 PDT 2026
Author: Jiahao Guo
Date: 2026-05-10T16:56:18+01:00
New Revision: 9c95f3765930b964d2a42c3f08dae2fd281cc49b
URL: https://github.com/llvm/llvm-project/commit/9c95f3765930b964d2a42c3f08dae2fd281cc49b
DIFF: https://github.com/llvm/llvm-project/commit/9c95f3765930b964d2a42c3f08dae2fd281cc49b.diff
LOG: [CIR][AArch64] Lower NEON vuzp intrinsics (#195591)
### Summary
part of : https://github.com/llvm/llvm-project/issues/185382
lower `vuzp` intrinsics in:
https://arm-software.github.io/acle/neon_intrinsics/advsimd.html#unzip-elements
this is a follow up : https://github.com/llvm/llvm-project/pull/195527
Lower `NEON::BI__builtin_neon_vuzp_v` and
`NEON::BI__builtin_neon_vuzpq_v`in CIRGenBuiltinAArch64.cpp by porting
by porting the existing incubator
logic(clangir/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp) : two
bitcasts on the input vectors,two rounds of cir.vec.shuffle generating
the deinterleave (even/odd) shuffle patterns with indices 2*i+vi, each
stored via ptr_stride on the sret base pointer.
Added:
Modified:
clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_untyped.c
clang/test/CodeGen/AArch64/neon-perm.c
clang/test/CodeGen/AArch64/neon/perm.c
Removed:
################################################################################
diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
index 16018d9b080ad..26c64c6b26dca 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -2871,12 +2871,30 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr,
case NEON::BI__builtin_neon_vst4q_lane_v:
case NEON::BI__builtin_neon_vtrn_v:
case NEON::BI__builtin_neon_vtrnq_v:
- case NEON::BI__builtin_neon_vuzp_v:
- case NEON::BI__builtin_neon_vuzpq_v:
cgm.errorNYI(expr->getSourceRange(),
std::string("unimplemented AArch64 builtin call: ") +
getContext().BuiltinInfo.getName(builtinID));
return mlir::Value{};
+ case NEON::BI__builtin_neon_vuzp_v:
+ case NEON::BI__builtin_neon_vuzpq_v: {
+ ops[1] = builder.createBitcast(ops[1], ty);
+ ops[2] = builder.createBitcast(ops[2], ty);
+ // Adding a bitcast here as Ops[0] might be a void pointer.
+ mlir::Value baseAddr =
+ builder.createBitcast(ops[0], builder.getPointerTo(ty));
+ mlir::Value sv;
+ for (unsigned vi = 0; vi != 2; ++vi) {
+ llvm::SmallVector<int64_t, 16> indices;
+ for (unsigned i = 0, e = ty.getSize(); i != e; ++i) {
+ indices.push_back(2 * i + vi);
+ }
+ cir::ConstantOp idx = builder.getConstInt(loc, builder.getSInt32Ty(), vi);
+ mlir::Value addr = builder.createPtrStride(loc, baseAddr, idx);
+ sv = builder.createVecShuffle(loc, ops[1], ops[2], indices);
+ (void)builder.CIRBaseBuilderTy::createStore(loc, sv, addr);
+ }
+ return sv;
+ }
case NEON::BI__builtin_neon_vzip_v:
case NEON::BI__builtin_neon_vzipq_v: {
ops[1] = builder.createBitcast(ops[1], ty);
diff --git a/clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_untyped.c b/clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_untyped.c
index bdb2cc1984963..0a9a88fc249b0 100644
--- a/clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_untyped.c
+++ b/clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_untyped.c
@@ -519,42 +519,6 @@ mfloat8x16x2_t test_vtrnq_mf8(mfloat8x16_t a, mfloat8x16_t b) {
return vtrnq_mf8(a, b);
}
-// CHECK-LABEL: define dso_local %struct.mfloat8x8x2_t @test_vuzp_mf8(
-// CHECK-SAME: <8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T:%.*]] poison, <8 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_0_INSERT1]], <8 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T]] poison, <8 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-mfloat8x8x2_t test_vuzp_mf8(mfloat8x8_t a, mfloat8x8_t b) {
- return vuzp_mf8(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.mfloat8x16x2_t @test_vuzpq_mf8(
-// CHECK-SAME: <16 x i8> [[A:%.*]], <16 x i8> [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T:%.*]] poison, <16 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_0_INSERT1]], <16 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T]] poison, <16 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-mfloat8x16x2_t test_vuzpq_mf8(mfloat8x16_t a, mfloat8x16_t b) {
- return vuzpq_mf8(a, b);
-}
-
// CHECK-LABEL: define dso_local void @test_vcopy_lane_mf8(
// CHECK-SAME: <8 x i8> [[ARG_I8X8:%.*]]) #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
diff --git a/clang/test/CodeGen/AArch64/neon-perm.c b/clang/test/CodeGen/AArch64/neon-perm.c
index edabc8a002953..7e3745b952f50 100644
--- a/clang/test/CodeGen/AArch64/neon-perm.c
+++ b/clang/test/CodeGen/AArch64/neon-perm.c
@@ -846,382 +846,6 @@ poly16x8_t test_vtrn2q_p16(poly16x8_t a, poly16x8_t b) {
return vtrn2q_p16(a, b);
}
-// CHECK-LABEL: define dso_local %struct.int8x8x2_t @test_vuzp_s8(
-// CHECK-SAME: <8 x i8> noundef [[A:%.*]], <8 x i8> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_INT8X8X2_T:%.*]] poison, <8 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_INT8X8X2_T]] [[DOTFCA_0_0_INSERT1]], <8 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_INT8X8X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_INT8X8X2_T]] poison, <8 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_INT8X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_INT8X8X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-int8x8x2_t test_vuzp_s8(int8x8_t a, int8x8_t b) {
- return vuzp_s8(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.int16x4x2_t @test_vuzp_s16(
-// CHECK-SAME: <4 x i16> noundef [[A:%.*]], <4 x i16> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i16> [[A]] to <8 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i16> [[B]] to <8 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <4 x i16> [[TMP2]], <4 x i16> [[TMP3]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <4 x i16> [[TMP2]], <4 x i16> [[TMP3]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_INT16X4X2_T:%.*]] poison, <4 x i16> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_INT16X4X2_T]] [[DOTFCA_0_0_INSERT1]], <4 x i16> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_INT16X4X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <4 x i16>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <4 x i16>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_INT16X4X2_T]] poison, <4 x i16> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_INT16X4X2_T]] [[DOTFCA_0_0_INSERT]], <4 x i16> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_INT16X4X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-int16x4x2_t test_vuzp_s16(int16x4_t a, int16x4_t b) {
- return vuzp_s16(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.int32x2x2_t @test_vuzp_s32(
-// CHECK-SAME: <2 x i32> noundef [[A:%.*]], <2 x i32> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x i32> [[A]] to <8 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i32> [[B]] to <8 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <2 x i32>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <2 x i32> [[TMP2]], <2 x i32> [[TMP3]], <2 x i32> <i32 0, i32 2>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <2 x i32> [[TMP2]], <2 x i32> [[TMP3]], <2 x i32> <i32 1, i32 3>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_INT32X2X2_T:%.*]] poison, <2 x i32> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_INT32X2X2_T]] [[DOTFCA_0_0_INSERT1]], <2 x i32> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_INT32X2X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <2 x i32>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <2 x i32>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_INT32X2X2_T]] poison, <2 x i32> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_INT32X2X2_T]] [[DOTFCA_0_0_INSERT]], <2 x i32> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_INT32X2X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-int32x2x2_t test_vuzp_s32(int32x2_t a, int32x2_t b) {
- return vuzp_s32(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.uint8x8x2_t @test_vuzp_u8(
-// CHECK-SAME: <8 x i8> noundef [[A:%.*]], <8 x i8> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_UINT8X8X2_T:%.*]] poison, <8 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_UINT8X8X2_T]] [[DOTFCA_0_0_INSERT1]], <8 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_UINT8X8X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_UINT8X8X2_T]] poison, <8 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_UINT8X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_UINT8X8X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-uint8x8x2_t test_vuzp_u8(uint8x8_t a, uint8x8_t b) {
- return vuzp_u8(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.uint16x4x2_t @test_vuzp_u16(
-// CHECK-SAME: <4 x i16> noundef [[A:%.*]], <4 x i16> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i16> [[A]] to <8 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i16> [[B]] to <8 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <4 x i16> [[TMP2]], <4 x i16> [[TMP3]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <4 x i16> [[TMP2]], <4 x i16> [[TMP3]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_UINT16X4X2_T:%.*]] poison, <4 x i16> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_UINT16X4X2_T]] [[DOTFCA_0_0_INSERT1]], <4 x i16> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_UINT16X4X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <4 x i16>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <4 x i16>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_UINT16X4X2_T]] poison, <4 x i16> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_UINT16X4X2_T]] [[DOTFCA_0_0_INSERT]], <4 x i16> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_UINT16X4X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-uint16x4x2_t test_vuzp_u16(uint16x4_t a, uint16x4_t b) {
- return vuzp_u16(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.uint32x2x2_t @test_vuzp_u32(
-// CHECK-SAME: <2 x i32> noundef [[A:%.*]], <2 x i32> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x i32> [[A]] to <8 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i32> [[B]] to <8 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <2 x i32>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <2 x i32> [[TMP2]], <2 x i32> [[TMP3]], <2 x i32> <i32 0, i32 2>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <2 x i32> [[TMP2]], <2 x i32> [[TMP3]], <2 x i32> <i32 1, i32 3>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_UINT32X2X2_T:%.*]] poison, <2 x i32> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_UINT32X2X2_T]] [[DOTFCA_0_0_INSERT1]], <2 x i32> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_UINT32X2X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <2 x i32>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <2 x i32>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_UINT32X2X2_T]] poison, <2 x i32> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_UINT32X2X2_T]] [[DOTFCA_0_0_INSERT]], <2 x i32> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_UINT32X2X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-uint32x2x2_t test_vuzp_u32(uint32x2_t a, uint32x2_t b) {
- return vuzp_u32(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.float32x2x2_t @test_vuzp_f32(
-// CHECK-SAME: <2 x float> noundef [[A:%.*]], <2 x float> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x float> [[A]] to <2 x i32>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x float> [[B]] to <2 x i32>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <2 x i32> [[TMP0]] to <8 x i8>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <2 x i32> [[TMP1]] to <8 x i8>
-// CHECK-NEXT: [[TMP4:%.*]] = bitcast <8 x i8> [[TMP2]] to <2 x float>
-// CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i8> [[TMP3]] to <2 x float>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> [[TMP5]], <2 x i32> <i32 0, i32 2>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> [[TMP5]], <2 x i32> <i32 1, i32 3>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT3:%.*]] = insertvalue [[STRUCT_FLOAT32X2X2_T:%.*]] poison, <2 x float> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT4:%.*]] = insertvalue [[STRUCT_FLOAT32X2X2_T]] [[DOTFCA_0_0_INSERT3]], <2 x float> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP6:%.*]] = extractvalue [[STRUCT_FLOAT32X2X2_T]] [[DOTFCA_0_1_INSERT4]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <2 x float>] [[TMP6]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <2 x float>] [[TMP6]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_FLOAT32X2X2_T]] poison, <2 x float> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_FLOAT32X2X2_T]] [[DOTFCA_0_0_INSERT]], <2 x float> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_FLOAT32X2X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-float32x2x2_t test_vuzp_f32(float32x2_t a, float32x2_t b) {
- return vuzp_f32(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.poly8x8x2_t @test_vuzp_p8(
-// CHECK-SAME: <8 x i8> noundef [[A:%.*]], <8 x i8> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_POLY8X8X2_T:%.*]] poison, <8 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_POLY8X8X2_T]] [[DOTFCA_0_0_INSERT1]], <8 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_POLY8X8X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_POLY8X8X2_T]] poison, <8 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_POLY8X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_POLY8X8X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-poly8x8x2_t test_vuzp_p8(poly8x8_t a, poly8x8_t b) {
- return vuzp_p8(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.poly16x4x2_t @test_vuzp_p16(
-// CHECK-SAME: <4 x i16> noundef [[A:%.*]], <4 x i16> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i16> [[A]] to <8 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i16> [[B]] to <8 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x i16>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <4 x i16> [[TMP2]], <4 x i16> [[TMP3]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <4 x i16> [[TMP2]], <4 x i16> [[TMP3]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_POLY16X4X2_T:%.*]] poison, <4 x i16> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_POLY16X4X2_T]] [[DOTFCA_0_0_INSERT1]], <4 x i16> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_POLY16X4X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <4 x i16>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <4 x i16>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_POLY16X4X2_T]] poison, <4 x i16> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_POLY16X4X2_T]] [[DOTFCA_0_0_INSERT]], <4 x i16> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_POLY16X4X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-poly16x4x2_t test_vuzp_p16(poly16x4_t a, poly16x4_t b) {
- return vuzp_p16(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.int8x16x2_t @test_vuzpq_s8(
-// CHECK-SAME: <16 x i8> noundef [[A:%.*]], <16 x i8> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_INT8X16X2_T:%.*]] poison, <16 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_INT8X16X2_T]] [[DOTFCA_0_0_INSERT1]], <16 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_INT8X16X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_INT8X16X2_T]] poison, <16 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_INT8X16X2_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_INT8X16X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-int8x16x2_t test_vuzpq_s8(int8x16_t a, int8x16_t b) {
- return vuzpq_s8(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.int16x8x2_t @test_vuzpq_s16(
-// CHECK-SAME: <8 x i16> noundef [[A:%.*]], <8 x i16> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x i16> [[A]] to <16 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i16> [[B]] to <16 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <8 x i16> [[TMP2]], <8 x i16> [[TMP3]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <8 x i16> [[TMP2]], <8 x i16> [[TMP3]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_INT16X8X2_T:%.*]] poison, <8 x i16> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_INT16X8X2_T]] [[DOTFCA_0_0_INSERT1]], <8 x i16> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_INT16X8X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x i16>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x i16>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_INT16X8X2_T]] poison, <8 x i16> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_INT16X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i16> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_INT16X8X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-int16x8x2_t test_vuzpq_s16(int16x8_t a, int16x8_t b) {
- return vuzpq_s16(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.int32x4x2_t @test_vuzpq_s32(
-// CHECK-SAME: <4 x i32> noundef [[A:%.*]], <4 x i32> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i32> [[A]] to <16 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i32> [[B]] to <16 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <4 x i32>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP3]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP3]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_INT32X4X2_T:%.*]] poison, <4 x i32> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_INT32X4X2_T]] [[DOTFCA_0_0_INSERT1]], <4 x i32> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_INT32X4X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <4 x i32>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <4 x i32>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_INT32X4X2_T]] poison, <4 x i32> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_INT32X4X2_T]] [[DOTFCA_0_0_INSERT]], <4 x i32> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_INT32X4X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-int32x4x2_t test_vuzpq_s32(int32x4_t a, int32x4_t b) {
- return vuzpq_s32(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.uint8x16x2_t @test_vuzpq_u8(
-// CHECK-SAME: <16 x i8> noundef [[A:%.*]], <16 x i8> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_UINT8X16X2_T:%.*]] poison, <16 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_UINT8X16X2_T]] [[DOTFCA_0_0_INSERT1]], <16 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_UINT8X16X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_UINT8X16X2_T]] poison, <16 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_UINT8X16X2_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_UINT8X16X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-uint8x16x2_t test_vuzpq_u8(uint8x16_t a, uint8x16_t b) {
- return vuzpq_u8(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.uint16x8x2_t @test_vuzpq_u16(
-// CHECK-SAME: <8 x i16> noundef [[A:%.*]], <8 x i16> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x i16> [[A]] to <16 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i16> [[B]] to <16 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <8 x i16> [[TMP2]], <8 x i16> [[TMP3]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <8 x i16> [[TMP2]], <8 x i16> [[TMP3]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_UINT16X8X2_T:%.*]] poison, <8 x i16> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_UINT16X8X2_T]] [[DOTFCA_0_0_INSERT1]], <8 x i16> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_UINT16X8X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x i16>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x i16>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_UINT16X8X2_T]] poison, <8 x i16> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_UINT16X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i16> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_UINT16X8X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-uint16x8x2_t test_vuzpq_u16(uint16x8_t a, uint16x8_t b) {
- return vuzpq_u16(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.uint32x4x2_t @test_vuzpq_u32(
-// CHECK-SAME: <4 x i32> noundef [[A:%.*]], <4 x i32> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i32> [[A]] to <16 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x i32> [[B]] to <16 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <4 x i32>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP3]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP3]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_UINT32X4X2_T:%.*]] poison, <4 x i32> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_UINT32X4X2_T]] [[DOTFCA_0_0_INSERT1]], <4 x i32> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_UINT32X4X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <4 x i32>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <4 x i32>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_UINT32X4X2_T]] poison, <4 x i32> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_UINT32X4X2_T]] [[DOTFCA_0_0_INSERT]], <4 x i32> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_UINT32X4X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-uint32x4x2_t test_vuzpq_u32(uint32x4_t a, uint32x4_t b) {
- return vuzpq_u32(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.float32x4x2_t @test_vuzpq_f32(
-// CHECK-SAME: <4 x float> noundef [[A:%.*]], <4 x float> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[A]] to <4 x i32>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <4 x float> [[B]] to <4 x i32>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <4 x i32> [[TMP0]] to <16 x i8>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <4 x i32> [[TMP1]] to <16 x i8>
-// CHECK-NEXT: [[TMP4:%.*]] = bitcast <16 x i8> [[TMP2]] to <4 x float>
-// CHECK-NEXT: [[TMP5:%.*]] = bitcast <16 x i8> [[TMP3]] to <4 x float>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <4 x float> [[TMP4]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <4 x float> [[TMP4]], <4 x float> [[TMP5]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT3:%.*]] = insertvalue [[STRUCT_FLOAT32X4X2_T:%.*]] poison, <4 x float> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT4:%.*]] = insertvalue [[STRUCT_FLOAT32X4X2_T]] [[DOTFCA_0_0_INSERT3]], <4 x float> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP6:%.*]] = extractvalue [[STRUCT_FLOAT32X4X2_T]] [[DOTFCA_0_1_INSERT4]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <4 x float>] [[TMP6]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <4 x float>] [[TMP6]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_FLOAT32X4X2_T]] poison, <4 x float> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_FLOAT32X4X2_T]] [[DOTFCA_0_0_INSERT]], <4 x float> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_FLOAT32X4X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-float32x4x2_t test_vuzpq_f32(float32x4_t a, float32x4_t b) {
- return vuzpq_f32(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.poly8x16x2_t @test_vuzpq_p8(
-// CHECK-SAME: <16 x i8> noundef [[A:%.*]], <16 x i8> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_POLY8X16X2_T:%.*]] poison, <16 x i8> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_POLY8X16X2_T]] [[DOTFCA_0_0_INSERT1]], <16 x i8> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_POLY8X16X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <16 x i8>] [[TMP0]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_POLY8X16X2_T]] poison, <16 x i8> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_POLY8X16X2_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_POLY8X16X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-poly8x16x2_t test_vuzpq_p8(poly8x16_t a, poly8x16_t b) {
- return vuzpq_p8(a, b);
-}
-
-// CHECK-LABEL: define dso_local %struct.poly16x8x2_t @test_vuzpq_p16(
-// CHECK-SAME: <8 x i16> noundef [[A:%.*]], <8 x i16> noundef [[B:%.*]]) #[[ATTR0]] {
-// CHECK-NEXT: [[ENTRY:.*:]]
-// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x i16> [[A]] to <16 x i8>
-// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i16> [[B]] to <16 x i8>
-// CHECK-NEXT: [[TMP2:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
-// CHECK-NEXT: [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <8 x i16>
-// CHECK-NEXT: [[VUZP_I:%.*]] = shufflevector <8 x i16> [[TMP2]], <8 x i16> [[TMP3]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
-// CHECK-NEXT: [[VUZP1_I:%.*]] = shufflevector <8 x i16> [[TMP2]], <8 x i16> [[TMP3]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT1:%.*]] = insertvalue [[STRUCT_POLY16X8X2_T:%.*]] poison, <8 x i16> [[VUZP_I]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT2:%.*]] = insertvalue [[STRUCT_POLY16X8X2_T]] [[DOTFCA_0_0_INSERT1]], <8 x i16> [[VUZP1_I]], 0, 1
-// CHECK-NEXT: [[TMP4:%.*]] = extractvalue [[STRUCT_POLY16X8X2_T]] [[DOTFCA_0_1_INSERT2]], 0
-// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x i16>] [[TMP4]], 0
-// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x i16>] [[TMP4]], 1
-// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_POLY16X8X2_T]] poison, <8 x i16> [[DOTFCA_0_EXTRACT]], 0, 0
-// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_POLY16X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i16> [[DOTFCA_1_EXTRACT]], 0, 1
-// CHECK-NEXT: ret [[STRUCT_POLY16X8X2_T]] [[DOTFCA_0_1_INSERT]]
-//
-poly16x8x2_t test_vuzpq_p16(poly16x8_t a, poly16x8_t b) {
- return vuzpq_p16(a, b);
-}
-
// CHECK-LABEL: define dso_local %struct.int8x8x2_t @test_vtrn_s8(
// CHECK-SAME: <8 x i8> noundef [[A:%.*]], <8 x i8> noundef [[B:%.*]]) #[[ATTR0]] {
// CHECK-NEXT: [[ENTRY:.*:]]
diff --git a/clang/test/CodeGen/AArch64/neon/perm.c b/clang/test/CodeGen/AArch64/neon/perm.c
index 6173ce43d18c7..6c3a9a44a7084 100644
--- a/clang/test/CodeGen/AArch64/neon/perm.c
+++ b/clang/test/CodeGen/AArch64/neon/perm.c
@@ -921,3 +921,385 @@ mfloat8x16x2_t test_vzipq_mf8(mfloat8x16_t a, mfloat8x16_t b) {
// LLVM: ret %struct.mfloat8x16x2_t
return vzipq_mf8(a, b);
}
+
+//===------------------------------------------------------===//
+// 2.1.9.11. Unzip elements
+// https://arm-software.github.io/acle/neon_intrinsics/advsimd.html#unzip-elements
+//===------------------------------------------------------===//
+
+// LLVM-LABEL: @test_vuzp_s8(
+// CIR-LABEL: @vuzp_s8(
+int8x8x2_t test_vuzp_s8(int8x8_t a, int8x8_t b) {
+// CIR: [[LO:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<8 x !s8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i] : !cir.vector<8 x !s8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<8 x !s8i>, !cir.ptr<!cir.vector<8 x !s8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<8 x !s8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i] : !cir.vector<8 x !s8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<8 x !s8i>, !cir.ptr<!cir.vector<8 x !s8i>>
+
+// LLVM-SAME: <8 x i8> {{.*}} [[A:%.*]], <8 x i8> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+// LLVM: insertvalue %struct.int8x8x2_t poison, <8 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.int8x8x2_t {{.*}}, <8 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.int8x8x2_t
+return vuzp_s8(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_s16(
+// CIR-LABEL: @vuzp_s16(
+int16x4x2_t test_vuzp_s16(int16x4_t a, int16x4_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<4 x !s16i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<4 x !s16i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !s16i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i] : !cir.vector<4 x !s16i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<4 x !s16i>, !cir.ptr<!cir.vector<4 x !s16i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !s16i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i] : !cir.vector<4 x !s16i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<4 x !s16i>, !cir.ptr<!cir.vector<4 x !s16i>>
+
+// LLVM-SAME: <4 x i16> {{.*}} [[A:%.*]], <4 x i16> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+// LLVM: insertvalue %struct.int16x4x2_t poison, <4 x i16> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.int16x4x2_t {{.*}}, <4 x i16> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.int16x4x2_t
+return vuzp_s16(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_u8(
+// CIR-LABEL: @vuzp_u8(
+uint8x8x2_t test_vuzp_u8(uint8x8_t a, uint8x8_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<8 x !u8i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<8 x !u8i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !u8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i] : !cir.vector<8 x !u8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<8 x !u8i>, !cir.ptr<!cir.vector<8 x !u8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !u8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i] : !cir.vector<8 x !u8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<8 x !u8i>, !cir.ptr<!cir.vector<8 x !u8i>>
+
+// LLVM-SAME: <8 x i8> {{.*}} [[A:%.*]], <8 x i8> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+// LLVM: insertvalue %struct.uint8x8x2_t poison, <8 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.uint8x8x2_t {{.*}}, <8 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.uint8x8x2_t
+return vuzp_u8(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_u16(
+// CIR-LABEL: @vuzp_u16(
+uint16x4x2_t test_vuzp_u16(uint16x4_t a, uint16x4_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<4 x !u16i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<4 x !u16i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !u16i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i] : !cir.vector<4 x !u16i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<4 x !u16i>, !cir.ptr<!cir.vector<4 x !u16i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !u16i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i] : !cir.vector<4 x !u16i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<4 x !u16i>, !cir.ptr<!cir.vector<4 x !u16i>>
+
+// LLVM-SAME: <4 x i16> {{.*}} [[A:%.*]], <4 x i16> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+// LLVM: insertvalue %struct.uint16x4x2_t poison, <4 x i16> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.uint16x4x2_t {{.*}}, <4 x i16> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.uint16x4x2_t
+return vuzp_u16(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_p8(
+// CIR-LABEL: @vuzp_p8(
+poly8x8x2_t test_vuzp_p8(poly8x8_t a, poly8x8_t b) {
+// CIR: [[LO:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<8 x !s8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i] : !cir.vector<8 x !s8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<8 x !s8i>, !cir.ptr<!cir.vector<8 x !s8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<8 x !s8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i] : !cir.vector<8 x !s8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<8 x !s8i>, !cir.ptr<!cir.vector<8 x !s8i>>
+
+// LLVM-SAME: <8 x i8> {{.*}} [[A:%.*]], <8 x i8> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+// LLVM: insertvalue %struct.poly8x8x2_t poison, <8 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.poly8x8x2_t {{.*}}, <8 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.poly8x8x2_t
+return vuzp_p8(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_p16(
+// CIR-LABEL: @vuzp_p16(
+poly16x4x2_t test_vuzp_p16(poly16x4_t a, poly16x4_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<4 x !s16i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<4 x !s16i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !s16i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i] : !cir.vector<4 x !s16i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<4 x !s16i>, !cir.ptr<!cir.vector<4 x !s16i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !s16i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i] : !cir.vector<4 x !s16i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<4 x !s16i>, !cir.ptr<!cir.vector<4 x !s16i>>
+
+// LLVM-SAME: <4 x i16> {{.*}} [[A:%.*]], <4 x i16> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+// LLVM: insertvalue %struct.poly16x4x2_t poison, <4 x i16> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.poly16x4x2_t {{.*}}, <4 x i16> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.poly16x4x2_t
+return vuzp_p16(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_mf8(
+// CIR-LABEL: @vuzp_mf8(
+mfloat8x8x2_t test_vuzp_mf8(mfloat8x8_t a, mfloat8x8_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<8 x !u8i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<8 x !u8i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !u8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i] : !cir.vector<8 x !u8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<8 x !u8i>, !cir.ptr<!cir.vector<8 x !u8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !u8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i] : !cir.vector<8 x !u8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<8 x !u8i>, !cir.ptr<!cir.vector<8 x !u8i>>
+
+// LLVM-SAME: <8 x i8> {{.*}}[[A:%.*]], <8 x i8> {{.*}}[[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+// LLVM: insertvalue %struct.mfloat8x8x2_t poison, <8 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.mfloat8x8x2_t {{.*}}, <8 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.mfloat8x8x2_t
+return vuzp_mf8(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_s32(
+// CIR-LABEL: @vuzp_s32(
+int32x2x2_t test_vuzp_s32(int32x2_t a, int32x2_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<2 x !s32i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<2 x !s32i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<2 x !s32i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i] : !cir.vector<2 x !s32i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<2 x !s32i>, !cir.ptr<!cir.vector<2 x !s32i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<2 x !s32i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i] : !cir.vector<2 x !s32i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<2 x !s32i>, !cir.ptr<!cir.vector<2 x !s32i>>
+
+// LLVM-SAME: <2 x i32> {{.*}} [[A:%.*]], <2 x i32> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> [[B]], <2 x i32> <i32 0, i32 2>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> [[B]], <2 x i32> <i32 1, i32 3>
+// LLVM: insertvalue %struct.int32x2x2_t poison, <2 x i32> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.int32x2x2_t {{.*}}, <2 x i32> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.int32x2x2_t
+return vuzp_s32(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_f32(
+// CIR-LABEL: @vuzp_f32(
+float32x2x2_t test_vuzp_f32(float32x2_t a, float32x2_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<2 x !cir.float>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<2 x !cir.float>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<2 x !cir.float>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i] : !cir.vector<2 x !cir.float>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<2 x !cir.float>, !cir.ptr<!cir.vector<2 x !cir.float>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<2 x !cir.float>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i] : !cir.vector<2 x !cir.float>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<2 x !cir.float>, !cir.ptr<!cir.vector<2 x !cir.float>>
+
+// LLVM-SAME: <2 x float> {{.*}} [[A:%.*]], <2 x float> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <2 x float> [[A]], <2 x float> [[B]], <2 x i32> <i32 0, i32 2>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <2 x float> [[A]], <2 x float> [[B]], <2 x i32> <i32 1, i32 3>
+// LLVM: insertvalue %struct.float32x2x2_t poison, <2 x float> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.float32x2x2_t {{.*}}, <2 x float> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.float32x2x2_t
+return vuzp_f32(a, b);
+}
+
+// LLVM-LABEL: @test_vuzp_u32(
+// CIR-LABEL: @vuzp_u32(
+uint32x2x2_t test_vuzp_u32(uint32x2_t a, uint32x2_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<2 x !u32i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<8 x !s8i> -> !cir.vector<2 x !u32i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<2 x !u32i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i] : !cir.vector<2 x !u32i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<2 x !u32i>, !cir.ptr<!cir.vector<2 x !u32i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<2 x !u32i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i] : !cir.vector<2 x !u32i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<2 x !u32i>, !cir.ptr<!cir.vector<2 x !u32i>>
+
+// LLVM-SAME: <2 x i32> {{.*}} [[A:%.*]], <2 x i32> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> [[B]], <2 x i32> <i32 0, i32 2>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> [[B]], <2 x i32> <i32 1, i32 3>
+// LLVM: insertvalue %struct.uint32x2x2_t poison, <2 x i32> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.uint32x2x2_t {{.*}}, <2 x i32> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.uint32x2x2_t
+return vuzp_u32(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_s8(
+// CIR-LABEL: @vuzpq_s8(
+int8x16x2_t test_vuzpq_s8(int8x16_t a, int8x16_t b) {
+// CIR: [[LO:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<16 x !s8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i, #cir.int<16> : !s32i, #cir.int<18> : !s32i, #cir.int<20> : !s32i, #cir.int<22> : !s32i, #cir.int<24> : !s32i, #cir.int<26> : !s32i, #cir.int<28> : !s32i, #cir.int<30> : !s32i] : !cir.vector<16 x !s8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<16 x !s8i>, !cir.ptr<!cir.vector<16 x !s8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<16 x !s8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i, #cir.int<17> : !s32i, #cir.int<19> : !s32i, #cir.int<21> : !s32i, #cir.int<23> : !s32i, #cir.int<25> : !s32i, #cir.int<27> : !s32i, #cir.int<29> : !s32i, #cir.int<31> : !s32i] : !cir.vector<16 x !s8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<16 x !s8i>, !cir.ptr<!cir.vector<16 x !s8i>>
+
+// LLVM-SAME: <16 x i8> {{.*}} [[A:%.*]], <16 x i8> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+// LLVM: insertvalue %struct.int8x16x2_t poison, <16 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.int8x16x2_t {{.*}}, <16 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.int8x16x2_t
+return vuzpq_s8(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_s16(
+// CIR-LABEL: @vuzpq_s16(
+int16x8x2_t test_vuzpq_s16(int16x8_t a, int16x8_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<8 x !s16i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<8 x !s16i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !s16i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i] : !cir.vector<8 x !s16i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<8 x !s16i>, !cir.ptr<!cir.vector<8 x !s16i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !s16i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i] : !cir.vector<8 x !s16i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<8 x !s16i>, !cir.ptr<!cir.vector<8 x !s16i>>
+
+// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]], <8 x i16> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+// LLVM: insertvalue %struct.int16x8x2_t poison, <8 x i16> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.int16x8x2_t {{.*}}, <8 x i16> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.int16x8x2_t
+return vuzpq_s16(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_s32(
+// CIR-LABEL: @vuzpq_s32(
+int32x4x2_t test_vuzpq_s32(int32x4_t a, int32x4_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<4 x !s32i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<4 x !s32i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !s32i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i] : !cir.vector<4 x !s32i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<4 x !s32i>, !cir.ptr<!cir.vector<4 x !s32i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !s32i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i] : !cir.vector<4 x !s32i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<4 x !s32i>, !cir.ptr<!cir.vector<4 x !s32i>>
+
+// LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]], <4 x i32> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+// LLVM: insertvalue %struct.int32x4x2_t poison, <4 x i32> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.int32x4x2_t {{.*}}, <4 x i32> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.int32x4x2_t
+return vuzpq_s32(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_f32(
+// CIR-LABEL: @vuzpq_f32(
+float32x4x2_t test_vuzpq_f32(float32x4_t a, float32x4_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<4 x !cir.float>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<4 x !cir.float>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !cir.float>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i] : !cir.vector<4 x !cir.float>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !cir.float>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i] : !cir.vector<4 x !cir.float>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<4 x !cir.float>, !cir.ptr<!cir.vector<4 x !cir.float>>
+
+// LLVM-SAME: <4 x float> {{.*}} [[A:%.*]], <4 x float> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <4 x float> [[A]], <4 x float> [[B]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <4 x float> [[A]], <4 x float> [[B]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+// LLVM: insertvalue %struct.float32x4x2_t poison, <4 x float> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.float32x4x2_t {{.*}}, <4 x float> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.float32x4x2_t
+return vuzpq_f32(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_u8(
+// CIR-LABEL: @vuzpq_u8(
+uint8x16x2_t test_vuzpq_u8(uint8x16_t a, uint8x16_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<16 x !u8i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<16 x !u8i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<16 x !u8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i, #cir.int<16> : !s32i, #cir.int<18> : !s32i, #cir.int<20> : !s32i, #cir.int<22> : !s32i, #cir.int<24> : !s32i, #cir.int<26> : !s32i, #cir.int<28> : !s32i, #cir.int<30> : !s32i] : !cir.vector<16 x !u8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<16 x !u8i>, !cir.ptr<!cir.vector<16 x !u8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<16 x !u8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i, #cir.int<17> : !s32i, #cir.int<19> : !s32i, #cir.int<21> : !s32i, #cir.int<23> : !s32i, #cir.int<25> : !s32i, #cir.int<27> : !s32i, #cir.int<29> : !s32i, #cir.int<31> : !s32i] : !cir.vector<16 x !u8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<16 x !u8i>, !cir.ptr<!cir.vector<16 x !u8i>>
+
+// LLVM-SAME: <16 x i8> {{.*}} [[A:%.*]], <16 x i8> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+// LLVM: insertvalue %struct.uint8x16x2_t poison, <16 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.uint8x16x2_t {{.*}}, <16 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.uint8x16x2_t
+return vuzpq_u8(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_u32(
+// CIR-LABEL: @vuzpq_u32(
+uint32x4x2_t test_vuzpq_u32(uint32x4_t a, uint32x4_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<4 x !u32i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<4 x !u32i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !u32i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i] : !cir.vector<4 x !u32i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<4 x !u32i>, !cir.ptr<!cir.vector<4 x !u32i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<4 x !u32i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i] : !cir.vector<4 x !u32i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<4 x !u32i>, !cir.ptr<!cir.vector<4 x !u32i>>
+
+// LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]], <4 x i32> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+// LLVM: insertvalue %struct.uint32x4x2_t poison, <4 x i32> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.uint32x4x2_t {{.*}}, <4 x i32> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.uint32x4x2_t
+return vuzpq_u32(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_p8(
+// CIR-LABEL: @vuzpq_p8(
+poly8x16x2_t test_vuzpq_p8(poly8x16_t a, poly8x16_t b) {
+// CIR: [[LO:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<16 x !s8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i, #cir.int<16> : !s32i, #cir.int<18> : !s32i, #cir.int<20> : !s32i, #cir.int<22> : !s32i, #cir.int<24> : !s32i, #cir.int<26> : !s32i, #cir.int<28> : !s32i, #cir.int<30> : !s32i] : !cir.vector<16 x !s8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<16 x !s8i>, !cir.ptr<!cir.vector<16 x !s8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle(%{{.*}}, %{{.*}} : !cir.vector<16 x !s8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i, #cir.int<17> : !s32i, #cir.int<19> : !s32i, #cir.int<21> : !s32i, #cir.int<23> : !s32i, #cir.int<25> : !s32i, #cir.int<27> : !s32i, #cir.int<29> : !s32i, #cir.int<31> : !s32i] : !cir.vector<16 x !s8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<16 x !s8i>, !cir.ptr<!cir.vector<16 x !s8i>>
+
+// LLVM-SAME: <16 x i8> {{.*}} [[A:%.*]], <16 x i8> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+// LLVM: insertvalue %struct.poly8x16x2_t poison, <16 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.poly8x16x2_t {{.*}}, <16 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.poly8x16x2_t
+return vuzpq_p8(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_p16(
+// CIR-LABEL: @vuzpq_p16(
+poly16x8x2_t test_vuzpq_p16(poly16x8_t a, poly16x8_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<8 x !s16i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<8 x !s16i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !s16i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i] : !cir.vector<8 x !s16i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<8 x !s16i>, !cir.ptr<!cir.vector<8 x !s16i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !s16i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i] : !cir.vector<8 x !s16i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<8 x !s16i>, !cir.ptr<!cir.vector<8 x !s16i>>
+
+// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]], <8 x i16> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+// LLVM: insertvalue %struct.poly16x8x2_t poison, <8 x i16> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.poly16x8x2_t {{.*}}, <8 x i16> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.poly16x8x2_t
+return vuzpq_p16(a, b);
+}
+
+// LLVM-LABEL: @test_vuzpq_mf8(
+// CIR-LABEL: @vuzpq_mf8(
+mfloat8x16x2_t test_vuzpq_mf8(mfloat8x16_t a, mfloat8x16_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<16 x !u8i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<16 x !u8i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<16 x !u8i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i, #cir.int<16> : !s32i, #cir.int<18> : !s32i, #cir.int<20> : !s32i, #cir.int<22> : !s32i, #cir.int<24> : !s32i, #cir.int<26> : !s32i, #cir.int<28> : !s32i, #cir.int<30> : !s32i] : !cir.vector<16 x !u8i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<16 x !u8i>, !cir.ptr<!cir.vector<16 x !u8i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<16 x !u8i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i, #cir.int<17> : !s32i, #cir.int<19> : !s32i, #cir.int<21> : !s32i, #cir.int<23> : !s32i, #cir.int<25> : !s32i, #cir.int<27> : !s32i, #cir.int<29> : !s32i, #cir.int<31> : !s32i] : !cir.vector<16 x !u8i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<16 x !u8i>, !cir.ptr<!cir.vector<16 x !u8i>>
+
+// LLVM-SAME: <16 x i8> {{.*}}[[A:%.*]], <16 x i8> {{.*}}[[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 16, i32 18, i32 20, i32 22, i32 24, i32 26, i32 28, i32 30>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
+// LLVM: insertvalue %struct.mfloat8x16x2_t poison, <16 x i8> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.mfloat8x16x2_t {{.*}}, <16 x i8> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.mfloat8x16x2_t
+return vuzpq_mf8(a, b);
+}
+
+//===------------------------------------------------------===//
+// 2.1.9.14. Unzip elements`
+// https://arm-software.github.io/acle/neon_intrinsics/advsimd.html#unzip-elements-1
+//===------------------------------------------------------===//
+
+// LLVM-LABEL: @test_vuzpq_u16(
+// CIR-LABEL: @vuzpq_u16(
+uint16x8x2_t test_vuzpq_u16(uint16x8_t a, uint16x8_t b) {
+// CIR: [[A_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<8 x !u16i>
+// CIR: [[B_CAST:%.*]] = cir.cast bitcast %{{.*}} : !cir.vector<16 x !s8i> -> !cir.vector<8 x !u16i>
+// CIR: [[LO:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !u16i>) [#cir.int<0> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<14> : !s32i] : !cir.vector<8 x !u16i>
+// CIR: cir.store [[LO]], %{{.*}} : !cir.vector<8 x !u16i>, !cir.ptr<!cir.vector<8 x !u16i>>
+// CIR: [[HI:%.*]] = cir.vec.shuffle([[A_CAST]], [[B_CAST]] : !cir.vector<8 x !u16i>) [#cir.int<1> : !s32i, #cir.int<3> : !s32i, #cir.int<5> : !s32i, #cir.int<7> : !s32i, #cir.int<9> : !s32i, #cir.int<11> : !s32i, #cir.int<13> : !s32i, #cir.int<15> : !s32i] : !cir.vector<8 x !u16i>
+// CIR: cir.store [[HI]], %{{.*}} : !cir.vector<8 x !u16i>, !cir.ptr<!cir.vector<8 x !u16i>>
+
+// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]], <8 x i16> {{.*}} [[B:%.*]]) {{.*}} {
+// LLVM: [[VUZP_LO:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
+// LLVM: [[VUZP_HI:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
+// LLVM: insertvalue %struct.uint16x8x2_t poison, <8 x i16> [[VUZP_LO]], 0, 0
+// LLVM: insertvalue %struct.uint16x8x2_t {{.*}}, <8 x i16> [[VUZP_HI]], 0, 1
+// LLVM: ret %struct.uint16x8x2_t
+return vuzpq_u16(a, b);
+}
More information about the cfe-commits
mailing list