This is an archive of the discontinued LLVM Phabricator instance.

Differential D112610

[LoopVectorize] Change getRuntimeVFAsFloat to use unsigned int->FP conversion
ClosedPublic

Authored by david-arm on Oct 27 2021, 3:45 AM.

Details

Reviewers
sdesmalen
kmclaughlin
bsmith
RosieSumpter
Commits
rG87a294d5eb5a: [LoopVectorize] Change getRuntimeVFAsFloat to use unsigned int->FP conversion
Summary

We never expect the runtime VF to be negative so we should use
the uitofp instruction instead of sitofp.

Diff Detail

Event Timeline

david-arm created this revision.Oct 27 2021, 3:45 AM
Herald added a subscriber: hiraditya. · View Herald TranscriptOct 27 2021, 3:45 AM
david-arm requested review of this revision.Oct 27 2021, 3:45 AM
Herald added a project: Restricted Project. · View Herald TranscriptOct 27 2021, 3:45 AM
Herald added a subscriber: llvm-commits. · View Herald Transcript
david-arm added a parent revision: D111882: [NFC][LoopVectorize] Change getStepVector to take a Value* for the StartIdx.Oct 27 2021, 3:45 AM
Harbormaster completed remote builds in B130908: Diff 382593.Oct 27 2021, 3:45 AM
kmclaughlin accepted this revision.Oct 27 2021, 3:50 AM
This revision is now accepted and ready to land.Oct 27 2021, 3:50 AM
This revision was landed with ongoing or failed builds.Nov 1 2021, 2:58 AM
Closed by commit rG87a294d5eb5a: [LoopVectorize] Change getRuntimeVFAsFloat to use unsigned int->FP conversion (authored by david-arm). · Explain Why
This revision was automatically updated to reflect the committed changes.
david-arm added a commit: rG87a294d5eb5a: [LoopVectorize] Change getRuntimeVFAsFloat to use unsigned int->FP conversion.

Revision Contents

PathSize
llvm/
lib/
Transforms/
Vectorize/
2 lines
test/
Transforms/
LoopVectorize/
2 lines

Diff 383755

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 1,117 Lines • ▼ Show 20 LinesValue *getRuntimeVF(IRBuilder<> &B, Type *Ty, ElementCount VF) {
Constant *EC = ConstantInt::get(Ty, VF.getKnownMinValue()); Constant *EC = ConstantInt::get(Ty, VF.getKnownMinValue());
return VF.isScalable() ? B.CreateVScale(EC) : EC; return VF.isScalable() ? B.CreateVScale(EC) : EC;
} }
static Value *getRuntimeVFAsFloat(IRBuilder<> &B, Type *FTy, ElementCount VF) { static Value *getRuntimeVFAsFloat(IRBuilder<> &B, Type *FTy, ElementCount VF) {
assert(FTy->isFloatingPointTy() && "Expected floating point type!"); assert(FTy->isFloatingPointTy() && "Expected floating point type!");
Type *IntTy = IntegerType::get(FTy->getContext(), FTy->getScalarSizeInBits()); Type *IntTy = IntegerType::get(FTy->getContext(), FTy->getScalarSizeInBits());
Value *RuntimeVF = getRuntimeVF(B, IntTy, VF); Value *RuntimeVF = getRuntimeVF(B, IntTy, VF);
return B.CreateSIToFP(RuntimeVF, FTy); return B.CreateUIToFP(RuntimeVF, FTy);
} }
void reportVectorizationFailure(const StringRef DebugMsg, void reportVectorizationFailure(const StringRef DebugMsg,
const StringRef OREMsg, const StringRef ORETag, const StringRef OREMsg, const StringRef ORETag,
OptimizationRemarkEmitter *ORE, Loop *TheLoop, OptimizationRemarkEmitter *ORE, Loop *TheLoop,
Instruction *I) { Instruction *I) {
LLVM_DEBUG(debugVectorizationMessage("Not vectorizing: ", DebugMsg, I)); LLVM_DEBUG(debugVectorizationMessage("Not vectorizing: ", DebugMsg, I));
LoopVectorizeHints Hints(TheLoop, true /* doesn't matter */, *ORE); LoopVectorizeHints Hints(TheLoop, true /* doesn't matter */, *ORE);
▲ Show 20 Lines • Show All 9,502 Lines • Show Last 20 Lines

llvm/test/Transforms/LoopVectorize/scalable-inductions.ll

Show First 20 Lines • Show All 140 Lines • ▼ Show 20 Lines
; CHECK-LABEL: @add_unique_indf32( ; CHECK-LABEL: @add_unique_indf32(
; CHECK: vector.ph: ; CHECK: vector.ph:
; CHECK: %[[STEPVEC:.*]] = call <vscale x 4 x i32> @llvm.experimental.stepvector.nxv4i32() ; CHECK: %[[STEPVEC:.*]] = call <vscale x 4 x i32> @llvm.experimental.stepvector.nxv4i32()
; CHECK-NEXT: %[[TMP1:.*]] = uitofp <vscale x 4 x i32> %[[STEPVEC]] to <vscale x 4 x float> ; CHECK-NEXT: %[[TMP1:.*]] = uitofp <vscale x 4 x i32> %[[STEPVEC]] to <vscale x 4 x float>
; CHECK-NEXT: %[[TMP2:.*]] = fmul <vscale x 4 x float> %[[TMP1]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 2.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer) ; CHECK-NEXT: %[[TMP2:.*]] = fmul <vscale x 4 x float> %[[TMP1]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 2.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer)
; CHECK-NEXT: %[[INDINIT:.*]] = fadd <vscale x 4 x float> %[[TMP2]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 0.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer) ; CHECK-NEXT: %[[INDINIT:.*]] = fadd <vscale x 4 x float> %[[TMP2]], shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 0.000000e+00, i32 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer)
; CHECK-NEXT: %[[VSCALE:.*]] = call i32 @llvm.vscale.i32() ; CHECK-NEXT: %[[VSCALE:.*]] = call i32 @llvm.vscale.i32()
; CHECK-NEXT: %[[TMP3:.*]] = shl i32 %8, 2 ; CHECK-NEXT: %[[TMP3:.*]] = shl i32 %8, 2
; CHECK-NEXT: %[[TMP4:.*]] = sitofp i32 %[[TMP3]] to float ; CHECK-NEXT: %[[TMP4:.*]] = uitofp i32 %[[TMP3]] to float
; CHECK-NEXT: %[[INC:.*]] = fmul float %[[TMP4]], 2.000000e+00 ; CHECK-NEXT: %[[INC:.*]] = fmul float %[[TMP4]], 2.000000e+00
; CHECK-NEXT: %[[TMP5:.*]] = insertelement <vscale x 4 x float> poison, float %[[INC]], i32 0 ; CHECK-NEXT: %[[TMP5:.*]] = insertelement <vscale x 4 x float> poison, float %[[INC]], i32 0
; CHECK-NEXT: %[[VECINC:.*]] = shufflevector <vscale x 4 x float> %[[TMP5]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer ; CHECK-NEXT: %[[VECINC:.*]] = shufflevector <vscale x 4 x float> %[[TMP5]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
; CHECK: vector.body: ; CHECK: vector.body:
; CHECK: %[[VECIND:.*]] = phi <vscale x 4 x float> [ %[[INDINIT]], %vector.ph ], [ %[[VECINDNXT:.*]], %vector.body ] ; CHECK: %[[VECIND:.*]] = phi <vscale x 4 x float> [ %[[INDINIT]], %vector.ph ], [ %[[VECINDNXT:.*]], %vector.body ]
; CHECK: store <vscale x 4 x float> %[[VECIND]] ; CHECK: store <vscale x 4 x float> %[[VECIND]]
; CHECK: %[[VECINDNXT]] = fadd <vscale x 4 x float> %[[VECIND]], %[[VECINC]] ; CHECK: %[[VECINDNXT]] = fadd <vscale x 4 x float> %[[VECIND]], %[[VECINC]]
Show All 28 Lines