Skip to content

[VectorCombine] Enable transform 'scalarizeLoadExtract' for scalable vector types #65443

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Sep 18, 2023
+18 −19
Merged

[VectorCombine] Enable transform 'scalarizeLoadExtract' for scalable vector types #65443

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 12 additions & 13 deletions llvm/lib/Transforms/Vectorize/VectorCombine.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1134,14 +1134,14 @@ bool VectorCombine::scalarizeLoadExtract(Instruction &I) {
if (!match(&I, m_Load(m_Value(Ptr))))
return false;

auto *FixedVT = cast<FixedVectorType>(I.getType());
auto *VecTy = cast<VectorType>(I.getType());
auto *LI = cast<LoadInst>(&I);
const DataLayout &DL = I.getModule()->getDataLayout();
if (LI->isVolatile() || !DL.typeSizeEqualsStoreSize(FixedVT))
if (LI->isVolatile() || !DL.typeSizeEqualsStoreSize(VecTy))
return false;

InstructionCost OriginalCost =
TTI.getMemoryOpCost(Instruction::Load, FixedVT, LI->getAlign(),
TTI.getMemoryOpCost(Instruction::Load, VecTy, LI->getAlign(),
LI->getPointerAddressSpace());
InstructionCost ScalarizedCost = 0;

Expand Down Expand Up @@ -1172,7 +1172,7 @@ bool VectorCombine::scalarizeLoadExtract(Instruction &I) {
LastCheckedInst = UI;
}

auto ScalarIdx = canScalarizeAccess(FixedVT, UI->getOperand(1), &I, AC, DT);
auto ScalarIdx = canScalarizeAccess(VecTy, UI->getOperand(1), &I, AC, DT);
if (!ScalarIdx.isSafe()) {
// TODO: Freeze index if it is safe to do so.
ScalarIdx.discard();
Expand All @@ -1182,12 +1182,12 @@ bool VectorCombine::scalarizeLoadExtract(Instruction &I) {
auto *Index = dyn_cast<ConstantInt>(UI->getOperand(1));
TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
OriginalCost +=
TTI.getVectorInstrCost(Instruction::ExtractElement, FixedVT, CostKind,
TTI.getVectorInstrCost(Instruction::ExtractElement, VecTy, CostKind,
Index ? Index->getZExtValue() : -1);
ScalarizedCost +=
TTI.getMemoryOpCost(Instruction::Load, FixedVT->getElementType(),
TTI.getMemoryOpCost(Instruction::Load, VecTy->getElementType(),
Align(1), LI->getPointerAddressSpace());
ScalarizedCost += TTI.getAddressComputationCost(FixedVT->getElementType());
ScalarizedCost += TTI.getAddressComputationCost(VecTy->getElementType());
}

if (ScalarizedCost >= OriginalCost)
Expand All @@ -1200,12 +1200,12 @@ bool VectorCombine::scalarizeLoadExtract(Instruction &I) {

Value *Idx = EI->getOperand(1);
Value *GEP =
Builder.CreateInBoundsGEP(FixedVT, Ptr, {Builder.getInt32(0), Idx});
Builder.CreateInBoundsGEP(VecTy, Ptr, {Builder.getInt32(0), Idx});
auto *NewLoad = cast<LoadInst>(Builder.CreateLoad(
FixedVT->getElementType(), GEP, EI->getName() + ".scalar"));
VecTy->getElementType(), GEP, EI->getName() + ".scalar"));

Align ScalarOpAlignment = computeAlignmentAfterScalarization(
LI->getAlign(), FixedVT->getElementType(), Idx, DL);
LI->getAlign(), VecTy->getElementType(), Idx, DL);
NewLoad->setAlignment(ScalarOpAlignment);

replaceValue(*EI, *NewLoad);
Expand Down Expand Up @@ -1727,9 +1727,6 @@ bool VectorCombine::run() {
case Instruction::ShuffleVector:
MadeChange |= widenSubvectorLoad(I);
break;
case Instruction::Load:
MadeChange |= scalarizeLoadExtract(I);
break;
default:
break;
}
Expand All @@ -1743,6 +1740,8 @@ bool VectorCombine::run() {
if (Opcode == Instruction::Store)
MadeChange |= foldSingleElementStore(I);

if (isa<VectorType>(I.getType()) && Opcode == Instruction::Load)
MadeChange |= scalarizeLoadExtract(I);

// If this is an early pipeline invocation of this pass, we are done.
if (TryEarlyFoldsOnly)
Expand Down
12 changes: 6 additions & 6 deletions llvm/test/Transforms/VectorCombine/AArch64/load-extractelement-scalarization.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,8 +15,8 @@ define i32 @load_extract_idx_0(ptr %x) {

define i32 @vscale_load_extract_idx_0(ptr %x) {
; CHECK-LABEL: @vscale_load_extract_idx_0(
; CHECK-NEXT: [[LV:%.*]] = load <vscale x 4 x i32>, ptr [[X:%.*]], align 16
; CHECK-NEXT: [[R:%.*]] = extractelement <vscale x 4 x i32> [[LV]], i32 0
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds <vscale x 4 x i32>, ptr [[X:%.*]], i32 0, i32 0
; CHECK-NEXT: [[R:%.*]] = load i32, ptr [[TMP1]], align 16
; CHECK-NEXT: ret i32 [[R]]
;
%lv = load <vscale x 4 x i32>, ptr %x
Expand Down Expand Up @@ -61,8 +61,8 @@ define i32 @load_extract_idx_2(ptr %x) {

define i32 @vscale_load_extract_idx_2(ptr %x) {
; CHECK-LABEL: @vscale_load_extract_idx_2(
; CHECK-NEXT: [[LV:%.*]] = load <vscale x 4 x i32>, ptr [[X:%.*]], align 16
; CHECK-NEXT: [[R:%.*]] = extractelement <vscale x 4 x i32> [[LV]], i32 2
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds <vscale x 4 x i32>, ptr [[X:%.*]], i32 0, i32 2
; CHECK-NEXT: [[R:%.*]] = load i32, ptr [[TMP1]], align 8
; CHECK-NEXT: ret i32 [[R]]
;
%lv = load <vscale x 4 x i32>, ptr %x
Expand Down Expand Up @@ -142,9 +142,9 @@ define i32 @vscale_load_extract_idx_var_i64_known_valid_by_assume(ptr %x, i64 %i
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[IDX:%.*]], 4
; CHECK-NEXT: call void @llvm.assume(i1 [[CMP]])
; CHECK-NEXT: [[LV:%.*]] = load <vscale x 4 x i32>, ptr [[X:%.*]], align 16
; CHECK-NEXT: call void @maythrow()
; CHECK-NEXT: [[R:%.*]] = extractelement <vscale x 4 x i32> [[LV]], i64 [[IDX]]
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds <vscale x 4 x i32>, ptr [[X:%.*]], i32 0, i64 [[IDX]]
; CHECK-NEXT: [[R:%.*]] = load i32, ptr [[TMP0]], align 4
; CHECK-NEXT: ret i32 [[R]]
;
entry:
Expand Down