Magic-Mirror/llvm-capstone
1
0
Fork 0
mirror of https://github.com/capstone-engine/llvm-capstone.git synced 2024-12-14 11:39:35 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[SROA] Create additional vector type candidates based on store and load slices

Browse Source 
Second try at A-Wadhwani's https://reviews.llvm.org/D132096, which was reverted.
The original patch had three issues:
* https://reviews.llvm.org/D134032, which bjope kindly fixed. That patch is merged into this one.
* [GHI #57796](https://github.com/llvm/llvm-project/issues/57796). Fixed and added a test.
* [GHI #57821](https://github.com/llvm/llvm-project/issues/57821). I believe this is an undefined behavior which is not the fault of the original patch. Please see the issue for more details.

Original diff summary:

This patch adds additional vector types to be considered when doing promotion in
SROA, based on the types of the store and load slices. This provides more
promotion opportunities, by potentially using an optimal "intermediate" vector
type.

For example, the following code would currently not be promoted to a vector,
since `__m128i` is a `<2 x i64>` vector.
```

__m128i packfoo0(int a, int b, int c, int d) {
  int r[4] = {a, b, c, d};
  __m128i rm;
  std::memcpy(&rm, r, sizeof(rm));
  return rm;
}
```
```
packfoo0(int, int, int, int):
  mov     dword ptr [rsp - 24], edi
  mov     dword ptr [rsp - 20], esi
  mov     dword ptr [rsp - 16], edx
  mov     dword ptr [rsp - 12], ecx
  movaps  xmm0, xmmword ptr [rsp - 24]
  ret
```
By also considering the types of the elements, we could find that the `<4 x i32>` type would be valid for promotion, hence removing the memory accesses for this function. In other words, we can explore other new vector types, with the same size but different element types based on the load and store instructions from the Slices, which can
provide us more promotion opportunities.

Additionally, the step for removing duplicate elements from the `CandidateTys` vector was not using an equality comparator, which has been fixed.

Differential Revision: https://reviews.llvm.org/D143225
This commit is contained in:
Han Zhu 2023-01-23 23:32:42 -08:00
parent d2b768bba3
commit f9c2a341b9
4 changed files with 132 additions and 70 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
llvm/lib/Transforms/Scalar/SROA.cpp
View File

@ -1976,6 +1976,7 @@ static VectorType *isVectorPromotionViable(Partition &P, const DataLayout &DL) {
// Collect the candidate types for vector-based promotion. Also track whether
// we have different element types.
SmallVector<VectorType *, 4> CandidateTys;
SetVector<Type *> LoadStoreTys;
Type *CommonEltTy = nullptr;
VectorType *CommonVecPtrTy = nullptr;
bool HaveVecPtrTy = false;
@ -2009,15 +2010,37 @@ static VectorType *isVectorPromotionViable(Partition &P, const DataLayout &DL) {
}
}
};
// Consider any loads or stores that are the exact size of the slice.
for (const Slice &S : P)
if (S.beginOffset() == P.beginOffset() &&
S.endOffset() == P.endOffset()) {
if (auto *LI = dyn_cast<LoadInst>(S.getUse()->getUser()))
CheckCandidateType(LI->getType());
else if (auto *SI = dyn_cast<StoreInst>(S.getUse()->getUser()))
CheckCandidateType(SI->getValueOperand()->getType());
// Put load and store types into a set for de-duplication.
for (const Slice &S : P) {
Type *Ty;
if (auto *LI = dyn_cast<LoadInst>(S.getUse()->getUser()))
Ty = LI->getType();
else if (auto *SI = dyn_cast<StoreInst>(S.getUse()->getUser()))
Ty = SI->getValueOperand()->getType();
else
continue;
LoadStoreTys.insert(Ty);
// Consider any loads or stores that are the exact size of the slice.
if (S.beginOffset() == P.beginOffset() && S.endOffset() == P.endOffset())
CheckCandidateType(Ty);
}
// Consider additional vector types where the element type size is a
// multiple of load/store element size.
for (Type *Ty : LoadStoreTys) {
if (!VectorType::isValidElementType(Ty))
continue;
unsigned TypeSize = DL.getTypeSizeInBits(Ty).getFixedValue();
for (VectorType *&VTy : CandidateTys) {
unsigned VectorSize = DL.getTypeSizeInBits(VTy).getFixedValue();
unsigned ElementSize =
DL.getTypeSizeInBits(VTy->getElementType()).getFixedValue();
if (TypeSize != VectorSize && TypeSize != ElementSize &&
VectorSize % TypeSize == 0) {
VectorType *NewVTy = VectorType::get(Ty, VectorSize / TypeSize, false);
CheckCandidateType(NewVTy);
}
}
}
// If we didn't find a vector type, nothing to do here.
if (CandidateTys.empty())

46
llvm/test/Transforms/SROA/pr57796.ll Normal file
View File

@ -0,0 +1,46 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes='sroa<preserve-cfg>' -S | FileCheck %s --check-prefixes=CHECK,CHECK-PRESERVE-CFG
; RUN: opt < %s -passes='sroa<modify-cfg>' -S | FileCheck %s --check-prefixes=CHECK,CHECK-MODIFY-CFG
%struct.Value = type { %union.anon }
%union.anon = type { <32 x i8> }
@A = dso_local global i64 0, align 8
; Make sure that sroa does not crash when dealing with an invalid vector
; element type.
define void @foo() {
; CHECK-LABEL: @foo(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[REF_TMP_I:%.*]] = alloca [[STRUCT_VALUE:%.*]], align 32
; CHECK-NEXT: call void @value_create(ptr sret([[STRUCT_VALUE]]) align 32 [[REF_TMP_I]])
; CHECK-NEXT: [[CALL_I:%.*]] = call align 32 ptr @value_set_type(ptr align 32 [[REF_TMP_I]])
; CHECK-NEXT: [[TMP0:%.*]] = load <32 x i8>, ptr [[CALL_I]], align 32
; CHECK-NEXT: [[REF_TMP_SROA_0_0_VEC_EXTRACT:%.*]] = shufflevector <32 x i8> [[TMP0]], <32 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i8> [[REF_TMP_SROA_0_0_VEC_EXTRACT]] to x86_mmx
; CHECK-NEXT: [[TMP2:%.*]] = call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx [[TMP1]], i8 0)
; CHECK-NEXT: store x86_mmx [[TMP2]], ptr @A, align 8
; CHECK-NEXT: ret void
;
entry:
%ref.tmp.i = alloca %struct.Value, align 32
%ref.tmp = alloca %struct.Value, align 32
call void @value_create(ptr sret(%struct.Value) align 32 %ref.tmp.i)
%call.i = call align 32 ptr @value_set_type(ptr align 32 %ref.tmp.i)
%0 = load <32 x i8>, ptr %call.i, align 32
store <32 x i8> %0, ptr %ref.tmp, align 32
%1 = load x86_mmx, ptr %ref.tmp, align 32
%2 = call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %1, i8 0)
store x86_mmx %2, ptr @A, align 8
ret void
}
declare x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx, i8 immarg)
declare dso_local void @value_create(ptr sret(%struct.Value) align 32)
declare dso_local align 32 ptr @value_set_type(ptr align 32)
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; CHECK-MODIFY-CFG: {{.*}}
; CHECK-PRESERVE-CFG: {{.*}}

35
llvm/test/Transforms/SROA/sroa-common-type-fail-promotion.ll
View File

@ -46,12 +46,15 @@ define amdgpu_kernel void @test_memset() #0 {
; CHECK-LABEL: @test_memset(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[DATA:%.*]] = load <4 x float>, ptr undef, align 16
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[DATA]] to <8 x half>
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[DATA]] to <8 x i16>
; CHECK-NEXT: br label [[BB:%.*]]
; CHECK: bb:
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 0
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 1
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 2
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 0
; CHECK-NEXT: [[TMP1:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT]] to half
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 1
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT]] to half
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 2
; CHECK-NEXT: [[TMP3:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT]] to half
; CHECK-NEXT: ret void
;
entry:
@ -217,7 +220,7 @@ define amdgpu_kernel void @test_struct_array_vector_i16() #0 {
; CHECK: bb:
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 0
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 1
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_4_0_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP1]], i32 0
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_4_16_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP1]], i32 0
; CHECK-NEXT: ret void
;
entry:
@ -285,12 +288,15 @@ define amdgpu_kernel void @test_array_vector() #0 {
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_5:%.*]] = alloca <8 x half>, align 16
; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 16 [[B_BLOCKWISE_COPY_SROA_5]], i8 0, i32 16, i1 false)
; CHECK-NEXT: [[DATA:%.*]] = load <4 x float>, ptr undef, align 16
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[DATA]] to <8 x half>
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[DATA]] to <8 x i16>
; CHECK-NEXT: br label [[BB:%.*]]
; CHECK: bb:
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 0
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 1
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 2
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 0
; CHECK-NEXT: [[TMP1:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT]] to half
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 1
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT]] to half
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 2
; CHECK-NEXT: [[TMP3:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT]] to half
; CHECK-NEXT: ret void
;
entry:
@ -315,12 +321,15 @@ define amdgpu_kernel void @test_array_vector2() #0 {
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_5:%.*]] = alloca <8 x half>, align 16
; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 16 [[B_BLOCKWISE_COPY_SROA_5]], i8 0, i32 16, i1 false)
; CHECK-NEXT: [[DATA:%.*]] = load <4 x float>, ptr undef, align 16
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[DATA]] to <8 x half>
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x float> [[DATA]] to <8 x i16>
; CHECK-NEXT: br label [[BB:%.*]]
; CHECK: bb:
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 0
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 1
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <8 x half> [[TMP0]], i32 2
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 0
; CHECK-NEXT: [[TMP1:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_0_VEC_EXTRACT]] to half
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 1
; CHECK-NEXT: [[TMP2:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_2_VEC_EXTRACT]] to half
; CHECK-NEXT: [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <8 x i16> [[TMP0]], i32 2
; CHECK-NEXT: [[TMP3:%.*]] = bitcast i16 [[B_BLOCKWISE_COPY_SROA_0_4_VEC_EXTRACT]] to half
; CHECK-NEXT: ret void
;
entry:

82
llvm/test/Transforms/SROA/vector-promotion.ll
View File

@ -891,34 +891,26 @@ define <2 x i64> @test13(i32 %a, i32 %b, i32 %c, i32 %d) {
; cast to a different vector type
; CHECK-LABEL: @test13(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[X_SROA_0:%.*]] = alloca <2 x i64>, align 16
; CHECK-NEXT: store i32 [[A:%.*]], ptr [[X_SROA_0]], align 16
; CHECK-NEXT: [[X_SROA_0_4_X_TMP2_SROA_IDX1:%.*]] = getelementptr inbounds i8, ptr [[X_SROA_0]], i64 4
; CHECK-NEXT: store i32 [[B:%.*]], ptr [[X_SROA_0_4_X_TMP2_SROA_IDX1]], align 4
; CHECK-NEXT: [[X_SROA_0_8_X_TMP3_SROA_IDX2:%.*]] = getelementptr inbounds i8, ptr [[X_SROA_0]], i64 8
; CHECK-NEXT: store i32 [[C:%.*]], ptr [[X_SROA_0_8_X_TMP3_SROA_IDX2]], align 8
; CHECK-NEXT: [[X_SROA_0_12_X_TMP4_SROA_IDX3:%.*]] = getelementptr inbounds i8, ptr [[X_SROA_0]], i64 12
; CHECK-NEXT: store i32 [[D:%.*]], ptr [[X_SROA_0_12_X_TMP4_SROA_IDX3]], align 4
; CHECK-NEXT: [[X_SROA_0_0_X_SROA_0_0_RESULT:%.*]] = load <2 x i64>, ptr [[X_SROA_0]], align 16
; CHECK-NEXT: ret <2 x i64> [[X_SROA_0_0_X_SROA_0_0_RESULT]]
; CHECK-NEXT: [[X_SROA_0_0_VEC_INSERT:%.*]] = insertelement <4 x i32> undef, i32 [[A:%.*]], i32 0
; CHECK-NEXT: [[X_SROA_0_4_VEC_INSERT:%.*]] = insertelement <4 x i32> [[X_SROA_0_0_VEC_INSERT]], i32 [[B:%.*]], i32 1
; CHECK-NEXT: [[X_SROA_0_8_VEC_INSERT:%.*]] = insertelement <4 x i32> [[X_SROA_0_4_VEC_INSERT]], i32 [[C:%.*]], i32 2
; CHECK-NEXT: [[X_SROA_0_12_VEC_INSERT:%.*]] = insertelement <4 x i32> [[X_SROA_0_8_VEC_INSERT]], i32 [[D:%.*]], i32 3
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i32> [[X_SROA_0_12_VEC_INSERT]] to <2 x i64>
; CHECK-NEXT: ret <2 x i64> [[TMP0]]
;
; DEBUG-LABEL: @test13(
; DEBUG-NEXT: entry:
; DEBUG-NEXT: [[X_SROA_0:%.*]] = alloca <2 x i64>, align 16, !dbg [[DBG354:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META349:![0-9]+]], metadata !DIExpression()), !dbg [[DBG354]]
; DEBUG-NEXT: store i32 [[A:%.*]], ptr [[X_SROA_0]], align 16, !dbg [[DBG355:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META349:![0-9]+]], metadata !DIExpression()), !dbg [[DBG354:![0-9]+]]
; DEBUG-NEXT: [[X_SROA_0_0_VEC_INSERT:%.*]] = insertelement <4 x i32> undef, i32 [[A:%.*]], i32 0, !dbg [[DBG355:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META350:![0-9]+]], metadata !DIExpression()), !dbg [[DBG356:![0-9]+]]
; DEBUG-NEXT: [[X_SROA_0_4_X_TMP2_SROA_IDX1:%.*]] = getelementptr inbounds i8, ptr [[X_SROA_0]], i64 4, !dbg [[DBG357:![0-9]+]]
; DEBUG-NEXT: store i32 [[B:%.*]], ptr [[X_SROA_0_4_X_TMP2_SROA_IDX1]], align 4, !dbg [[DBG357]]
; DEBUG-NEXT: [[X_SROA_0_4_VEC_INSERT:%.*]] = insertelement <4 x i32> [[X_SROA_0_0_VEC_INSERT]], i32 [[B:%.*]], i32 1, !dbg [[DBG357:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META351:![0-9]+]], metadata !DIExpression()), !dbg [[DBG358:![0-9]+]]
; DEBUG-NEXT: [[X_SROA_0_8_X_TMP3_SROA_IDX2:%.*]] = getelementptr inbounds i8, ptr [[X_SROA_0]], i64 8, !dbg [[DBG359:![0-9]+]]
; DEBUG-NEXT: store i32 [[C:%.*]], ptr [[X_SROA_0_8_X_TMP3_SROA_IDX2]], align 8, !dbg [[DBG359]]
; DEBUG-NEXT: [[X_SROA_0_8_VEC_INSERT:%.*]] = insertelement <4 x i32> [[X_SROA_0_4_VEC_INSERT]], i32 [[C:%.*]], i32 2, !dbg [[DBG359:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META352:![0-9]+]], metadata !DIExpression()), !dbg [[DBG360:![0-9]+]]
; DEBUG-NEXT: [[X_SROA_0_12_X_TMP4_SROA_IDX3:%.*]] = getelementptr inbounds i8, ptr [[X_SROA_0]], i64 12, !dbg [[DBG361:![0-9]+]]
; DEBUG-NEXT: store i32 [[D:%.*]], ptr [[X_SROA_0_12_X_TMP4_SROA_IDX3]], align 4, !dbg [[DBG361]]
; DEBUG-NEXT: [[X_SROA_0_0_X_SROA_0_0_RESULT:%.*]] = load <2 x i64>, ptr [[X_SROA_0]], align 16, !dbg [[DBG362:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata <2 x i64> [[X_SROA_0_0_X_SROA_0_0_RESULT]], metadata [[META353:![0-9]+]], metadata !DIExpression()), !dbg [[DBG362]]
; DEBUG-NEXT: ret <2 x i64> [[X_SROA_0_0_X_SROA_0_0_RESULT]], !dbg [[DBG363:![0-9]+]]
; DEBUG-NEXT: [[X_SROA_0_12_VEC_INSERT:%.*]] = insertelement <4 x i32> [[X_SROA_0_8_VEC_INSERT]], i32 [[D:%.*]], i32 3, !dbg [[DBG361:![0-9]+]]
; DEBUG-NEXT: [[TMP0:%.*]] = bitcast <4 x i32> [[X_SROA_0_12_VEC_INSERT]] to <2 x i64>, !dbg [[DBG362:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata <2 x i64> [[TMP0]], metadata [[META353:![0-9]+]], metadata !DIExpression()), !dbg [[DBG362]]
; DEBUG-NEXT: ret <2 x i64> [[TMP0]], !dbg [[DBG363:![0-9]+]]
;
entry:
%x = alloca [4 x i32]
@ -938,43 +930,35 @@ define i32 @test14(<2 x i64> %x) {
; cast to a different vector type
; CHECK-LABEL: @test14(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[X_ADDR:%.*]] = alloca <2 x i64>, align 16
; CHECK-NEXT: store <2 x i64> [[X:%.*]], ptr [[X_ADDR]], align 16
; CHECK-NEXT: [[X_ADDR_0_A:%.*]] = load i32, ptr [[X_ADDR]], align 16
; CHECK-NEXT: [[X_ADDR_4_X_TMP2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[X_ADDR]], i64 4
; CHECK-NEXT: [[X_ADDR_4_B:%.*]] = load i32, ptr [[X_ADDR_4_X_TMP2_SROA_IDX]], align 4
; CHECK-NEXT: [[X_ADDR_8_X_TMP3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[X_ADDR]], i64 8
; CHECK-NEXT: [[X_ADDR_8_C:%.*]] = load i32, ptr [[X_ADDR_8_X_TMP3_SROA_IDX]], align 8
; CHECK-NEXT: [[X_ADDR_12_X_TMP4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[X_ADDR]], i64 12
; CHECK-NEXT: [[X_ADDR_12_D:%.*]] = load i32, ptr [[X_ADDR_12_X_TMP4_SROA_IDX]], align 4
; CHECK-NEXT: [[ADD:%.*]] = add i32 [[X_ADDR_0_A]], [[X_ADDR_4_B]]
; CHECK-NEXT: [[ADD1:%.*]] = add i32 [[X_ADDR_8_C]], [[X_ADDR_12_D]]
; CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x i64> [[X:%.*]] to <4 x i32>
; CHECK-NEXT: [[X_ADDR_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 0
; CHECK-NEXT: [[X_ADDR_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 1
; CHECK-NEXT: [[X_ADDR_SROA_0_8_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 2
; CHECK-NEXT: [[X_ADDR_SROA_0_12_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
; CHECK-NEXT: [[ADD:%.*]] = add i32 [[X_ADDR_SROA_0_0_VEC_EXTRACT]], [[X_ADDR_SROA_0_4_VEC_EXTRACT]]
; CHECK-NEXT: [[ADD1:%.*]] = add i32 [[X_ADDR_SROA_0_8_VEC_EXTRACT]], [[X_ADDR_SROA_0_12_VEC_EXTRACT]]
; CHECK-NEXT: [[ADD2:%.*]] = add i32 [[ADD]], [[ADD1]]
; CHECK-NEXT: ret i32 [[ADD2]]
;
; DEBUG-LABEL: @test14(
; DEBUG-NEXT: entry:
; DEBUG-NEXT: [[X_ADDR:%.*]] = alloca <2 x i64>, align 16, !dbg [[DBG378:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr [[X_ADDR]], metadata [[META366:![0-9]+]], metadata !DIExpression()), !dbg [[DBG378]]
; DEBUG-NEXT: store <2 x i64> [[X:%.*]], ptr [[X_ADDR]], align 16, !dbg [[DBG379:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META366:![0-9]+]], metadata !DIExpression()), !dbg [[DBG378:![0-9]+]]
; DEBUG-NEXT: [[TMP0:%.*]] = bitcast <2 x i64> [[X:%.*]] to <4 x i32>, !dbg [[DBG379:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META367:![0-9]+]], metadata !DIExpression()), !dbg [[DBG380:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_0_A:%.*]] = load i32, ptr [[X_ADDR]], align 16, !dbg [[DBG381:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_0_A]], metadata [[META368:![0-9]+]], metadata !DIExpression()), !dbg [[DBG381]]
; DEBUG-NEXT: [[X_ADDR_SROA_0_0_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 0, !dbg [[DBG381:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_SROA_0_0_VEC_EXTRACT]], metadata [[META368:![0-9]+]], metadata !DIExpression()), !dbg [[DBG381]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META369:![0-9]+]], metadata !DIExpression()), !dbg [[DBG382:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_4_X_TMP2_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[X_ADDR]], i64 4, !dbg [[DBG383:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_4_B:%.*]] = load i32, ptr [[X_ADDR_4_X_TMP2_SROA_IDX]], align 4, !dbg [[DBG383]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_4_B]], metadata [[META370:![0-9]+]], metadata !DIExpression()), !dbg [[DBG383]]
; DEBUG-NEXT: [[X_ADDR_SROA_0_4_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 1, !dbg [[DBG383:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_SROA_0_4_VEC_EXTRACT]], metadata [[META370:![0-9]+]], metadata !DIExpression()), !dbg [[DBG383]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META371:![0-9]+]], metadata !DIExpression()), !dbg [[DBG384:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_8_X_TMP3_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[X_ADDR]], i64 8, !dbg [[DBG385:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_8_C:%.*]] = load i32, ptr [[X_ADDR_8_X_TMP3_SROA_IDX]], align 8, !dbg [[DBG385]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_8_C]], metadata [[META372:![0-9]+]], metadata !DIExpression()), !dbg [[DBG385]]
; DEBUG-NEXT: [[X_ADDR_SROA_0_8_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 2, !dbg [[DBG385:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_SROA_0_8_VEC_EXTRACT]], metadata [[META372:![0-9]+]], metadata !DIExpression()), !dbg [[DBG385]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata ptr undef, metadata [[META373:![0-9]+]], metadata !DIExpression()), !dbg [[DBG386:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_12_X_TMP4_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[X_ADDR]], i64 12, !dbg [[DBG387:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_12_D:%.*]] = load i32, ptr [[X_ADDR_12_X_TMP4_SROA_IDX]], align 4, !dbg [[DBG387]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_12_D]], metadata [[META374:![0-9]+]], metadata !DIExpression()), !dbg [[DBG387]]
; DEBUG-NEXT: [[ADD:%.*]] = add i32 [[X_ADDR_0_A]], [[X_ADDR_4_B]], !dbg [[DBG388:![0-9]+]]
; DEBUG-NEXT: [[X_ADDR_SROA_0_12_VEC_EXTRACT:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3, !dbg [[DBG387:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[X_ADDR_SROA_0_12_VEC_EXTRACT]], metadata [[META374:![0-9]+]], metadata !DIExpression()), !dbg [[DBG387]]
; DEBUG-NEXT: [[ADD:%.*]] = add i32 [[X_ADDR_SROA_0_0_VEC_EXTRACT]], [[X_ADDR_SROA_0_4_VEC_EXTRACT]], !dbg [[DBG388:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[ADD]], metadata [[META375:![0-9]+]], metadata !DIExpression()), !dbg [[DBG388]]
; DEBUG-NEXT: [[ADD1:%.*]] = add i32 [[X_ADDR_8_C]], [[X_ADDR_12_D]], !dbg [[DBG389:![0-9]+]]
; DEBUG-NEXT: [[ADD1:%.*]] = add i32 [[X_ADDR_SROA_0_8_VEC_EXTRACT]], [[X_ADDR_SROA_0_12_VEC_EXTRACT]], !dbg [[DBG389:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[ADD1]], metadata [[META376:![0-9]+]], metadata !DIExpression()), !dbg [[DBG389]]
; DEBUG-NEXT: [[ADD2:%.*]] = add i32 [[ADD]], [[ADD1]], !dbg [[DBG390:![0-9]+]]
; DEBUG-NEXT: call void @llvm.dbg.value(metadata i32 [[ADD2]], metadata [[META377:![0-9]+]], metadata !DIExpression()), !dbg [[DBG390]]