Skip to content

[VectorCombine] Apply InstSimplify in scalarizeOpOrCmp to avoid infinite loop #153069

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 4 commits into from
Aug 15, 2025
Merged

[VectorCombine] Apply InstSimplify in scalarizeOpOrCmp to avoid infinite loop #153069

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
42208a2
[VectorCombine] Precisely calculate the cost in foldInstExtBinop and …
XChy Aug 11, 2025
6a5666b
fix testcases
XChy Aug 12, 2025
716a5b1
fix comment and testcase
XChy Aug 14, 2025
2f15724
Use InstSimplify
XChy Aug 15, 2025
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
28 changes: 11 additions & 17 deletions llvm/lib/Transforms/Vectorize/VectorCombine.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,7 +74,7 @@ class VectorCombine {
const DataLayout *DL, TTI::TargetCostKind CostKind,
bool TryEarlyFoldsOnly)
: F(F), Builder(F.getContext(), InstSimplifyFolder(*DL)), TTI(TTI),
DT(DT), AA(AA), AC(AC), DL(DL), CostKind(CostKind),
DT(DT), AA(AA), AC(AC), DL(DL), CostKind(CostKind), SQ(*DL),
TryEarlyFoldsOnly(TryEarlyFoldsOnly) {}

bool run();
Expand All @@ -88,6 +88,7 @@ class VectorCombine {
AssumptionCache ∾
const DataLayout *DL;
TTI::TargetCostKind CostKind;
const SimplifyQuery SQ;

/// If true, only perform beneficial early IR transforms. Do not introduce new
/// vector operations.
Expand Down Expand Up @@ -1232,24 +1233,26 @@ bool VectorCombine::scalarizeOpOrCmp(Instruction &I) {
// Fold the vector constants in the original vectors into a new base vector to
// get more accurate cost modelling.
Value *NewVecC = nullptr;
TargetFolder Folder(*DL);
if (CI)
NewVecC = Folder.FoldCmp(CI->getPredicate(), VecCs[0], VecCs[1]);
NewVecC = simplifyCmpInst(CI->getPredicate(), VecCs[0], VecCs[1], SQ);
else if (UO)
NewVecC =
Folder.FoldUnOpFMF(UO->getOpcode(), VecCs[0], UO->getFastMathFlags());
simplifyUnOp(UO->getOpcode(), VecCs[0], UO->getFastMathFlags(), SQ);
else if (BO)
NewVecC = Folder.FoldBinOp(BO->getOpcode(), VecCs[0], VecCs[1]);
else if (II->arg_size() == 2)
NewVecC = Folder.FoldBinaryIntrinsic(II->getIntrinsicID(), VecCs[0],
VecCs[1], II->getType(), &I);
NewVecC = simplifyBinOp(BO->getOpcode(), VecCs[0], VecCs[1], SQ);
else if (II)
NewVecC = simplifyCall(II, II->getCalledOperand(), VecCs, SQ);

if (!NewVecC)
return false;

// Get cost estimate for the insert element. This cost will factor into
// both sequences.
InstructionCost OldCost = VectorOpCost;
InstructionCost NewCost =
ScalarOpCost + TTI.getVectorInstrCost(Instruction::InsertElement, VecTy,
CostKind, *Index, NewVecC);

Copy link
Contributor

@lukel97 lukel97 Aug 15, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry for the delay with the review.

After thinking about this for a bit, I think scalarizeOpOrCmp should only scalarize iff the base vector NewVecC can be folded.

I think that's what the original code in 0d2a0b4 kind of assumed. It's probably never profitable to do the scalarization if the base vector doesn't go away.

And whenever I was last working on scalarizeOpOrCmp, my only motivation was to scalarize scalable splats, which are inserts into poison vectors that are constant foldable anyway.

If we change TargetFolder to InstSimplifyFolder in this PR to minimize the regressions, and do something like

Suggested change
if (!NewVecC)
return nullptr;

then we can get rid of the early out in foldInsExtBinop, and also this bit below:

  // Create a new base vector if the constant folding failed.
  if (!NewVecC) {
    if (CI)
      NewVecC = Builder.CreateCmp(CI->getPredicate(), VecCs[0], VecCs[1]);
    else if (UO || BO)
      NewVecC = Builder.CreateNAryOp(Opcode, VecCs);
    else
      NewVecC = Builder.CreateIntrinsic(VecTy, II->getIntrinsicID(), VecCs);
  }

I think the only regression we would be left with is that we can't constant fold n-ary intrinsics at the moment, but this is something that we should fix in IRBuilderFolder. This would basically undo #138406, but I'm happy to take those regressions for now if we want this fix to land in for LLVM 21.

What do you think? cc @RKSimon

Copy link
Contributor

@lukel97 lukel97 Aug 15, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've opened up #153743 to track folding n-ary intrinsics

Copy link
Member Author

@XChy XChy Aug 15, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It sounds reasonable to me if there is no possible profitability. And I think simplifyInstruction(Instruction *I, const SimplifyQuery &Q) in InstSimplify would resolve your concern about n-ary intrinsics.

Copy link
Member Author

@XChy XChy Aug 15, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Or ConstantFoldInstOperands may be simpler.

Copy link
Member Author

@XChy XChy Aug 15, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

simplifyInstWithOperands requires distinguishing intrinsics among the operators. Thus, I implement by simplifyCmpInst, simplifyUnOp, etc., case by case.

Copy link
Contributor

@lukel97 lukel97 Aug 15, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh perfect, great to see we could avoid the regression. Thanks!

for (auto [Idx, Op, VecC, Scalar] : enumerate(Ops, VecCs, ScalarOps)) {
if (!Scalar || (II && isVectorIntrinsicWithScalarOpAtArg(
II->getIntrinsicID(), Idx, &TTI)))
Expand Down Expand Up @@ -1294,15 +1297,6 @@ bool VectorCombine::scalarizeOpOrCmp(Instruction &I) {
if (auto *ScalarInst = dyn_cast<Instruction>(Scalar))
ScalarInst->copyIRFlags(&I);

// Create a new base vector if the constant folding failed.
if (!NewVecC) {
if (CI)
NewVecC = Builder.CreateCmp(CI->getPredicate(), VecCs[0], VecCs[1]);
else if (UO || BO)
NewVecC = Builder.CreateNAryOp(Opcode, VecCs);
else
NewVecC = Builder.CreateIntrinsic(VecTy, II->getIntrinsicID(), VecCs);
}
Value *Insert = Builder.CreateInsertElement(NewVecC, Scalar, *Index);
replaceValue(I, *Insert);
return true;
Expand Down
3 changes: 1 addition & 2 deletions llvm/test/Transforms/VectorCombine/X86/intrinsic-scalarize.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,8 +13,7 @@ define <2 x float> @maxnum(float %x, float %y) {
; AVX2-LABEL: define <2 x float> @maxnum(
; AVX2-SAME: float [[X:%.*]], float [[Y:%.*]]) #[[ATTR0:[0-9]+]] {
; AVX2-NEXT: [[V_SCALAR:%.*]] = call float @llvm.maxnum.f32(float [[X]], float [[Y]])
; AVX2-NEXT: [[TMP1:%.*]] = call <2 x float> @llvm.maxnum.v2f32(<2 x float> poison, <2 x float> poison)
; AVX2-NEXT: [[V:%.*]] = insertelement <2 x float> [[TMP1]], float [[V_SCALAR]], i64 0
; AVX2-NEXT: [[V:%.*]] = insertelement <2 x float> poison, float [[V_SCALAR]], i64 0
; AVX2-NEXT: ret <2 x float> [[V]]
;
%x.insert = insertelement <2 x float> poison, float %x, i32 0
Expand Down
19 changes: 19 additions & 0 deletions llvm/test/Transforms/VectorCombine/binop-scalarize.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,3 +20,22 @@ define <4 x i8> @udiv_ub(i8 %x, i8 %y) {
%v = udiv <4 x i8> %x.insert, %y.insert
ret <4 x i8> %v
}


; Unfoldable constant expression may cause infinite loop between
; scalarizing insertelement and folding binop(insert(x,a,idx),insert(y,b,idx))
@val = external hidden global ptr, align 8

define <2 x i64> @pr153012(i64 %idx) #0 {
; CHECK-LABEL: define <2 x i64> @pr153012(
; CHECK-SAME: i64 [[IDX:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[A:%.*]] = insertelement <2 x i64> <i64 5, i64 ptrtoint (ptr @val to i64)>, i64 [[IDX]], i32 0
; CHECK-NEXT: [[B:%.*]] = or disjoint <2 x i64> splat (i64 2), [[A]]
; CHECK-NEXT: ret <2 x i64> [[B]]
;
entry:
%a = insertelement <2 x i64> <i64 5, i64 ptrtoint (ptr @val to i64)>, i64 %idx, i32 0
%b = or disjoint <2 x i64> splat (i64 2), %a
ret <2 x i64> %b
}
33 changes: 11 additions & 22 deletions llvm/test/Transforms/VectorCombine/intrinsic-scalarize.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,8 +5,7 @@ define <4 x i32> @umax_fixed(i32 %x, i32 %y) {
; CHECK-LABEL: define <4 x i32> @umax_fixed(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call i32 @llvm.umax.i32(i32 [[X]], i32 [[Y]])
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> @llvm.umax.v4i32(<4 x i32> poison, <4 x i32> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x i32> poison, i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <4 x i32> [[V]]
;
%x.insert = insertelement <4 x i32> poison, i32 %x, i32 0
Expand All @@ -19,8 +18,7 @@ define <vscale x 4 x i32> @umax_scalable(i32 %x, i32 %y) {
; CHECK-LABEL: define <vscale x 4 x i32> @umax_scalable(
; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call i32 @llvm.umax.i32(i32 [[X]], i32 [[Y]])
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x i32> @llvm.umax.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x i32> [[TMP1]], i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <vscale x 4 x i32> [[V]]
;
%x.insert = insertelement <vscale x 4 x i32> poison, i32 %x, i32 0
Expand All @@ -33,8 +31,7 @@ define <4 x i32> @umax_fixed_lhs_const(i32 %x) {
; CHECK-LABEL: define <4 x i32> @umax_fixed_lhs_const(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call i32 @llvm.umax.i32(i32 1, i32 [[X]])
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> @llvm.umax.v4i32(<4 x i32> <i32 1, i32 2, i32 3, i32 4>, <4 x i32> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x i32> poison, i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <4 x i32> [[V]]
;
%x.insert = insertelement <4 x i32> poison, i32 %x, i32 0
Expand All @@ -46,8 +43,7 @@ define <4 x i32> @umax_fixed_rhs_const(i32 %x) {
; CHECK-LABEL: define <4 x i32> @umax_fixed_rhs_const(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call i32 @llvm.umax.i32(i32 [[X]], i32 1)
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> @llvm.umax.v4i32(<4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 3, i32 4>)
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x i32> poison, i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <4 x i32> [[V]]
;
%x.insert = insertelement <4 x i32> poison, i32 %x, i32 0
Expand All @@ -59,8 +55,7 @@ define <vscale x 4 x i32> @umax_scalable_lhs_const(i32 %x) {
; CHECK-LABEL: define <vscale x 4 x i32> @umax_scalable_lhs_const(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call i32 @llvm.umax.i32(i32 42, i32 [[X]])
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x i32> @llvm.umax.nxv4i32(<vscale x 4 x i32> splat (i32 42), <vscale x 4 x i32> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x i32> [[TMP1]], i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <vscale x 4 x i32> [[V]]
;
%x.insert = insertelement <vscale x 4 x i32> poison, i32 %x, i32 0
Expand All @@ -72,8 +67,7 @@ define <vscale x 4 x i32> @umax_scalable_rhs_const(i32 %x) {
; CHECK-LABEL: define <vscale x 4 x i32> @umax_scalable_rhs_const(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call i32 @llvm.umax.i32(i32 [[X]], i32 42)
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x i32> @llvm.umax.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> splat (i32 42))
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x i32> [[TMP1]], i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <vscale x 4 x i32> [[V]]
;
%x.insert = insertelement <vscale x 4 x i32> poison, i32 %x, i32 0
Expand All @@ -100,8 +94,7 @@ define <4 x float> @fabs_fixed(float %x) {
; CHECK-LABEL: define <4 x float> @fabs_fixed(
; CHECK-SAME: float [[X:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x float> @llvm.fabs.v4f32(<4 x float> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x float> [[TMP1]], float [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x float> poison, float [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <4 x float> [[V]]
;
%x.insert = insertelement <4 x float> poison, float %x, i32 0
Expand All @@ -113,8 +106,7 @@ define <vscale x 4 x float> @fabs_scalable(float %x) {
; CHECK-LABEL: define <vscale x 4 x float> @fabs_scalable(
; CHECK-SAME: float [[X:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x float> @llvm.fabs.nxv4f32(<vscale x 4 x float> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x float> [[TMP1]], float [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x float> poison, float [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <vscale x 4 x float> [[V]]
;
%x.insert = insertelement <vscale x 4 x float> poison, float %x, i32 0
Expand All @@ -126,8 +118,7 @@ define <4 x float> @fma_fixed(float %x, float %y, float %z) {
; CHECK-LABEL: define <4 x float> @fma_fixed(
; CHECK-SAME: float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call float @llvm.fma.f32(float [[X]], float [[Y]], float [[Z]])
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x float> @llvm.fma.v4f32(<4 x float> poison, <4 x float> poison, <4 x float> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x float> [[TMP1]], float [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x float> poison, float [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <4 x float> [[V]]
;
%x.insert = insertelement <4 x float> poison, float %x, i32 0
Expand All @@ -141,8 +132,7 @@ define <vscale x 4 x float> @fma_scalable(float %x, float %y, float %z) {
; CHECK-LABEL: define <vscale x 4 x float> @fma_scalable(
; CHECK-SAME: float [[X:%.*]], float [[Y:%.*]], float [[Z:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call float @llvm.fma.f32(float [[X]], float [[Y]], float [[Z]])
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x float> @llvm.fma.nxv4f32(<vscale x 4 x float> poison, <vscale x 4 x float> poison, <vscale x 4 x float> poison)
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x float> [[TMP1]], float [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <vscale x 4 x float> poison, float [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <vscale x 4 x float> [[V]]
;
%x.insert = insertelement <vscale x 4 x float> poison, float %x, i32 0
Expand All @@ -156,8 +146,7 @@ define <4 x float> @scalar_argument(float %x) {
; CHECK-LABEL: define <4 x float> @scalar_argument(
; CHECK-SAME: float [[X:%.*]]) {
; CHECK-NEXT: [[V_SCALAR:%.*]] = call float @llvm.powi.f32.i32(float [[X]], i32 42)
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x float> @llvm.powi.v4f32.i32(<4 x float> poison, i32 42)
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x float> [[TMP1]], float [[V_SCALAR]], i64 0
; CHECK-NEXT: [[V:%.*]] = insertelement <4 x float> poison, float [[V_SCALAR]], i64 0
; CHECK-NEXT: ret <4 x float> [[V]]
;
%x.insert = insertelement <4 x float> poison, float %x, i32 0
Expand Down