Author: Craig Topper Date: 2025-05-02T16:10:18-07:00 New Revision: 123758b1f4a8106926d95268ea8dc27158b6393a
URL: https://github.com/llvm/llvm-project/commit/123758b1f4a8106926d95268ea8dc27158b6393a DIFF: https://github.com/llvm/llvm-project/commit/123758b1f4a8106926d95268ea8dc27158b6393a.diff LOG: [IRBuilder] Add versions of createInsertVector/createExtractVector that take a uint64_t index. (#138324) Most callers want a constant index. Instead of making every caller create a ConstantInt, we can do it in IRBuilder. This is similar to createInsertElement/createExtractElement. Added: Modified: clang/lib/CodeGen/CGCall.cpp clang/lib/CodeGen/CGExprScalar.cpp clang/lib/CodeGen/TargetBuiltins/ARM.cpp llvm/include/llvm/IR/IRBuilder.h llvm/lib/IR/AutoUpgrade.cpp llvm/lib/Target/AArch64/AArch64ISelLowering.cpp llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp Removed: ################################################################################ diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp index b5a7217dd3e56..82a24f7c295a2 100644 --- a/clang/lib/CodeGen/CGCall.cpp +++ b/clang/lib/CodeGen/CGCall.cpp @@ -1372,9 +1372,8 @@ static llvm::Value *CreateCoercedLoad(Address Src, llvm::Type *Ty, if (ScalableDstTy->getElementType() == FixedSrcTy->getElementType()) { auto *Load = CGF.Builder.CreateLoad(Src); auto *PoisonVec = llvm::PoisonValue::get(ScalableDstTy); - auto *Zero = llvm::Constant::getNullValue(CGF.CGM.Int64Ty); llvm::Value *Result = CGF.Builder.CreateInsertVector( - ScalableDstTy, PoisonVec, Load, Zero, "cast.scalable"); + ScalableDstTy, PoisonVec, Load, uint64_t(0), "cast.scalable"); if (ScalableDstTy != Ty) Result = CGF.Builder.CreateBitCast(Result, Ty); return Result; @@ -1482,10 +1481,8 @@ CoerceScalableToFixed(CodeGenFunction &CGF, llvm::FixedVectorType *ToTy, V = CGF.Builder.CreateBitCast(V, FromTy); } if (FromTy->getElementType() == ToTy->getElementType()) { - llvm::Value *Zero = llvm::Constant::getNullValue(CGF.CGM.Int64Ty); - V->setName(Name + ".coerce"); - V = CGF.Builder.CreateExtractVector(ToTy, V, Zero, "cast.fixed"); + V = CGF.Builder.CreateExtractVector(ToTy, V, uint64_t(0), "cast.fixed"); return {V, true}; } return {V, false}; @@ -6110,8 +6107,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo, dyn_cast<llvm::ScalableVectorType>(V->getType())) { if (FixedDstTy->getElementType() == ScalableSrcTy->getElementType()) { - llvm::Value *Zero = llvm::Constant::getNullValue(CGM.Int64Ty); - V = Builder.CreateExtractVector(FixedDstTy, V, Zero, + V = Builder.CreateExtractVector(FixedDstTy, V, uint64_t(0), "cast.fixed"); return RValue::get(V); } diff --git a/clang/lib/CodeGen/CGExprScalar.cpp b/clang/lib/CodeGen/CGExprScalar.cpp index 8dbbcdaef25d8..15a6177746403 100644 --- a/clang/lib/CodeGen/CGExprScalar.cpp +++ b/clang/lib/CodeGen/CGExprScalar.cpp @@ -2487,9 +2487,8 @@ Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) { } if (FixedSrcTy->getElementType() == ScalableDstTy->getElementType()) { llvm::Value *PoisonVec = llvm::PoisonValue::get(ScalableDstTy); - llvm::Value *Zero = llvm::Constant::getNullValue(CGF.CGM.Int64Ty); llvm::Value *Result = Builder.CreateInsertVector( - ScalableDstTy, PoisonVec, Src, Zero, "cast.scalable"); + ScalableDstTy, PoisonVec, Src, uint64_t(0), "cast.scalable"); if (Result->getType() != DstTy) Result = Builder.CreateBitCast(Result, DstTy); return Result; @@ -2512,10 +2511,9 @@ Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) { ScalableSrcTy->getElementCount().getKnownMinValue() / 8); Src = Builder.CreateBitCast(Src, ScalableSrcTy); } - if (ScalableSrcTy->getElementType() == FixedDstTy->getElementType()) { - llvm::Value *Zero = llvm::Constant::getNullValue(CGF.CGM.Int64Ty); - return Builder.CreateExtractVector(DstTy, Src, Zero, "cast.fixed"); - } + if (ScalableSrcTy->getElementType() == FixedDstTy->getElementType()) + return Builder.CreateExtractVector(DstTy, Src, uint64_t(0), + "cast.fixed"); } } diff --git a/clang/lib/CodeGen/TargetBuiltins/ARM.cpp b/clang/lib/CodeGen/TargetBuiltins/ARM.cpp index ec2e9722028f1..d37e68508373c 100644 --- a/clang/lib/CodeGen/TargetBuiltins/ARM.cpp +++ b/clang/lib/CodeGen/TargetBuiltins/ARM.cpp @@ -466,7 +466,7 @@ llvm::Value *CodeGenFunction::EmitFP8NeonFDOTCall( if (ExtendLaneArg) { auto *VT = llvm::FixedVectorType::get(Int8Ty, 16); Ops[2] = Builder.CreateInsertVector(VT, PoisonValue::get(VT), Ops[2], - Builder.getInt64(0)); + uint64_t(0)); } return EmitFP8NeonCall(IID, Tys, Ops, E, name); } @@ -478,7 +478,7 @@ llvm::Value *CodeGenFunction::EmitFP8NeonFMLACall( if (ExtendLaneArg) { auto *VT = llvm::FixedVectorType::get(Int8Ty, 16); Ops[2] = Builder.CreateInsertVector(VT, PoisonValue::get(VT), Ops[2], - Builder.getInt64(0)); + uint64_t(0)); } const unsigned ElemCount = Ops[0]->getType()->getPrimitiveSizeInBits() / RetTy->getPrimitiveSizeInBits(); @@ -502,7 +502,7 @@ Value *CodeGenFunction::EmitFP8NeonCvtCall(unsigned IID, llvm::Type *Ty0, // Op[0] is mfloat8x16_t, but the intrinsic converts only the lower part of // the vector. Tys[1] = llvm::FixedVectorType::get(Int8Ty, 8); - Ops[0] = Builder.CreateExtractVector(Tys[1], Ops[0], Builder.getInt64(0)); + Ops[0] = Builder.CreateExtractVector(Tys[1], Ops[0], uint64_t(0)); } return EmitFP8NeonCall(IID, Tys, Ops, E, name); } @@ -4727,7 +4727,7 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, llvm::Type *OverloadedTy = getSVEVectorForElementType(EltTy); Value *InsertSubVec = Builder.CreateInsertVector( - OverloadedTy, PoisonValue::get(OverloadedTy), Vec, Builder.getInt64(0)); + OverloadedTy, PoisonValue::get(OverloadedTy), Vec, uint64_t(0)); Function *F = CGM.getIntrinsic(Intrinsic::aarch64_sve_dupq_lane, OverloadedTy); @@ -4810,7 +4810,7 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, case SVE::BI__builtin_sve_svset_neonq_f32: case SVE::BI__builtin_sve_svset_neonq_f64: case SVE::BI__builtin_sve_svset_neonq_bf16: { - return Builder.CreateInsertVector(Ty, Ops[0], Ops[1], Builder.getInt64(0)); + return Builder.CreateInsertVector(Ty, Ops[0], Ops[1], uint64_t(0)); } case SVE::BI__builtin_sve_svget_neonq_s8: @@ -4825,7 +4825,7 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, case SVE::BI__builtin_sve_svget_neonq_f32: case SVE::BI__builtin_sve_svget_neonq_f64: case SVE::BI__builtin_sve_svget_neonq_bf16: { - return Builder.CreateExtractVector(Ty, Ops[0], Builder.getInt64(0)); + return Builder.CreateExtractVector(Ty, Ops[0], uint64_t(0)); } case SVE::BI__builtin_sve_svdup_neonq_s8: @@ -4841,7 +4841,7 @@ Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, case SVE::BI__builtin_sve_svdup_neonq_f64: case SVE::BI__builtin_sve_svdup_neonq_bf16: { Value *Insert = Builder.CreateInsertVector(Ty, PoisonValue::get(Ty), Ops[0], - Builder.getInt64(0)); + uint64_t(0)); return Builder.CreateIntrinsic(Intrinsic::aarch64_sve_dupq_lane, {Ty}, {Insert, Builder.getInt64(0)}); } @@ -7767,7 +7767,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID, case NEON::BI__builtin_neon_vcvt_high_mf8_f32_fpm: { llvm::Type *Ty = llvm::FixedVectorType::get(Int8Ty, 16); Ops[0] = Builder.CreateInsertVector(Ty, PoisonValue::get(Ty), Ops[0], - Builder.getInt64(0)); + uint64_t(0)); return EmitFP8NeonCvtCall(Intrinsic::aarch64_neon_fp8_fcvtn2, Ty, Ops[1]->getType(), false, Ops, E, "vfcvtn2"); } diff --git a/llvm/include/llvm/IR/IRBuilder.h b/llvm/include/llvm/IR/IRBuilder.h index 0e68ffadc6939..f650c06590ef2 100644 --- a/llvm/include/llvm/IR/IRBuilder.h +++ b/llvm/include/llvm/IR/IRBuilder.h @@ -1101,6 +1101,12 @@ class IRBuilderBase { Name); } + /// Create a call to the vector.extract intrinsic. + CallInst *CreateExtractVector(Type *DstType, Value *SrcVec, uint64_t Idx, + const Twine &Name = "") { + return CreateExtractVector(DstType, SrcVec, getInt64(Idx), Name); + } + /// Create a call to the vector.insert intrinsic. CallInst *CreateInsertVector(Type *DstType, Value *SrcVec, Value *SubVec, Value *Idx, const Twine &Name = "") { @@ -1109,6 +1115,12 @@ class IRBuilderBase { nullptr, Name); } + /// Create a call to the vector.extract intrinsic. + CallInst *CreateInsertVector(Type *DstType, Value *SrcVec, Value *SubVec, + uint64_t Idx, const Twine &Name = "") { + return CreateInsertVector(DstType, SrcVec, SubVec, getInt64(Idx), Name); + } + /// Create a call to llvm.stacksave CallInst *CreateStackSave(const Twine &Name = "") { const DataLayout &DL = BB->getDataLayout(); diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp index 2fd5b715b078f..8bf93555b1fd2 100644 --- a/llvm/lib/IR/AutoUpgrade.cpp +++ b/llvm/lib/IR/AutoUpgrade.cpp @@ -4554,9 +4554,8 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) { Value *NewLdCall = Builder.CreateCall(NewFn, Args); Value *Ret = llvm::PoisonValue::get(RetTy); for (unsigned I = 0; I < N; I++) { - Value *Idx = ConstantInt::get(Type::getInt64Ty(C), I * MinElts); Value *SRet = Builder.CreateExtractValue(NewLdCall, I); - Ret = Builder.CreateInsertVector(RetTy, Ret, SRet, Idx); + Ret = Builder.CreateInsertVector(RetTy, Ret, SRet, I * MinElts); } NewCall = dyn_cast<CallInst>(Ret); break; @@ -4611,9 +4610,8 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) { Value *Ret = llvm::PoisonValue::get(RetTy); unsigned MinElts = RetTy->getMinNumElements() / N; for (unsigned I = 0; I < N; I++) { - Value *Idx = ConstantInt::get(Type::getInt64Ty(C), I * MinElts); Value *V = CI->getArgOperand(I); - Ret = Builder.CreateInsertVector(RetTy, Ret, V, Idx); + Ret = Builder.CreateInsertVector(RetTy, Ret, V, I * MinElts); } NewCall = dyn_cast<CallInst>(Ret); } diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp index af4780e11e890..a62aa443983e1 100644 --- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp +++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp @@ -17227,9 +17227,7 @@ bool AArch64TargetLowering::lowerInterleavedLoad( Value *SubVec = Builder.CreateExtractValue(LdN, Index); if (UseScalable) - SubVec = Builder.CreateExtractVector( - FVTy, SubVec, - ConstantInt::get(Type::getInt64Ty(VTy->getContext()), 0)); + SubVec = Builder.CreateExtractVector(FVTy, SubVec, uint64_t(0)); // Convert the integer vector to pointer vector if the element is pointer. if (EltTy->isPointerTy()) @@ -17436,9 +17434,8 @@ bool AArch64TargetLowering::lowerInterleavedStore(StoreInst *SI, } if (UseScalable) - Shuffle = Builder.CreateInsertVector( - STVTy, PoisonValue::get(STVTy), Shuffle, - ConstantInt::get(Type::getInt64Ty(STVTy->getContext()), 0)); + Shuffle = Builder.CreateInsertVector(STVTy, PoisonValue::get(STVTy), + Shuffle, uint64_t(0)); Ops.push_back(Shuffle); } @@ -30020,31 +30017,26 @@ Value *AArch64TargetLowering::createComplexDeinterleavingIR( .getKnownMinValue() / 2; auto *HalfTy = VectorType::getHalfElementsVectorType(Ty); - auto *LowerSplitA = B.CreateExtractVector(HalfTy, InputA, B.getInt64(0)); - auto *LowerSplitB = B.CreateExtractVector(HalfTy, InputB, B.getInt64(0)); - auto *UpperSplitA = - B.CreateExtractVector(HalfTy, InputA, B.getInt64(Stride)); - auto *UpperSplitB = - B.CreateExtractVector(HalfTy, InputB, B.getInt64(Stride)); + auto *LowerSplitA = B.CreateExtractVector(HalfTy, InputA, uint64_t(0)); + auto *LowerSplitB = B.CreateExtractVector(HalfTy, InputB, uint64_t(0)); + auto *UpperSplitA = B.CreateExtractVector(HalfTy, InputA, Stride); + auto *UpperSplitB = B.CreateExtractVector(HalfTy, InputB, Stride); Value *LowerSplitAcc = nullptr; Value *UpperSplitAcc = nullptr; Type *FullTy = Ty; FullTy = Accumulator->getType(); auto *HalfAccTy = VectorType::getHalfElementsVectorType( cast<VectorType>(Accumulator->getType())); - LowerSplitAcc = - B.CreateExtractVector(HalfAccTy, Accumulator, B.getInt64(0)); - UpperSplitAcc = - B.CreateExtractVector(HalfAccTy, Accumulator, B.getInt64(AccStride)); + LowerSplitAcc = B.CreateExtractVector(HalfAccTy, Accumulator, uint64_t(0)); + UpperSplitAcc = B.CreateExtractVector(HalfAccTy, Accumulator, AccStride); auto *LowerSplitInt = createComplexDeinterleavingIR( B, OperationType, Rotation, LowerSplitA, LowerSplitB, LowerSplitAcc); auto *UpperSplitInt = createComplexDeinterleavingIR( B, OperationType, Rotation, UpperSplitA, UpperSplitB, UpperSplitAcc); auto *Result = B.CreateInsertVector(FullTy, PoisonValue::get(FullTy), - LowerSplitInt, B.getInt64(0)); - return B.CreateInsertVector(FullTy, Result, UpperSplitInt, - B.getInt64(AccStride)); + LowerSplitInt, uint64_t(0)); + return B.CreateInsertVector(FullTy, Result, UpperSplitInt, AccStride); } if (OperationType == ComplexDeinterleavingOperation::CMulPartial) { diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp index 05d3241cbe2f8..d99a30083fb88 100644 --- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp +++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp @@ -2406,9 +2406,9 @@ static std::optional<Instruction *> instCombineSVEUzp1(InstCombiner &IC, if (TyA == B->getType() && RetTy == ScalableVectorType::getDoubleElementsVectorType(TyA)) { auto *SubVec = IC.Builder.CreateInsertVector( - RetTy, PoisonValue::get(RetTy), A, IC.Builder.getInt64(0)); - auto *ConcatVec = IC.Builder.CreateInsertVector( - RetTy, SubVec, B, IC.Builder.getInt64(TyA->getMinNumElements())); + RetTy, PoisonValue::get(RetTy), A, uint64_t(0)); + auto *ConcatVec = IC.Builder.CreateInsertVector(RetTy, SubVec, B, + TyA->getMinNumElements()); ConcatVec->takeName(&II); return IC.replaceInstUsesWith(II, ConcatVec); } @@ -2602,9 +2602,9 @@ static std::optional<Instruction *> instCombineSVEDupqLane(InstCombiner &IC, auto *WideShuffleMaskTy = ScalableVectorType::get(IC.Builder.getInt32Ty(), PatternElementCount); - auto ZeroIdx = ConstantInt::get(IC.Builder.getInt64Ty(), APInt(64, 0)); auto InsertSubvector = IC.Builder.CreateInsertVector( - II.getType(), PoisonValue::get(II.getType()), InsertEltChain, ZeroIdx); + II.getType(), PoisonValue::get(II.getType()), InsertEltChain, + uint64_t(0)); auto WideBitcast = IC.Builder.CreateBitOrPointerCast(InsertSubvector, WideScalableTy); auto WideShuffleMask = ConstantAggregateZero::get(WideShuffleMaskTy); diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp index 3915028b557d7..efb2894aaf642 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp @@ -2102,8 +2102,7 @@ bool AMDGPUCodeGenPrepareImpl::visitPHINode(PHINode &I) { for (VectorSlice &S : Slices) { const auto ValName = "largephi.insertslice" + std::to_string(NameSuffix++); if (S.NumElts > 1) - Vec = - B.CreateInsertVector(FVT, Vec, S.NewPHI, B.getInt64(S.Idx), ValName); + Vec = B.CreateInsertVector(FVT, Vec, S.NewPHI, S.Idx, ValName); else Vec = B.CreateInsertElement(Vec, S.NewPHI, S.Idx, ValName); } diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp index 7ec2ee06b811a..8679a7cd0f88d 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp @@ -1440,14 +1440,14 @@ GCNTTIImpl::instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II) const { if (Src0Ty->getNumElements() > Src0NumElts) { Src0 = IC.Builder.CreateExtractVector( FixedVectorType::get(Src0Ty->getElementType(), Src0NumElts), Src0, - IC.Builder.getInt64(0)); + uint64_t(0)); MadeChange = true; } if (Src1Ty->getNumElements() > Src1NumElts) { Src1 = IC.Builder.CreateExtractVector( FixedVectorType::get(Src1Ty->getElementType(), Src1NumElts), Src1, - IC.Builder.getInt64(0)); + uint64_t(0)); MadeChange = true; } diff --git a/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp index 90329200dd7e4..02ffc67c774dd 100644 --- a/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp +++ b/llvm/lib/Transforms/Vectorize/LoopIdiomVectorize.cpp @@ -1322,9 +1322,8 @@ Value *LoopIdiomVectorize::expandFindFirstByte( Needle0, "needle0"); LoadNeedle = Builder.CreateSelect(PredNeedle, LoadNeedle, Needle0Splat, "needle_splat"); - LoadNeedle = - Builder.CreateExtractVector(FixedVectorType::get(CharTy, VF), LoadNeedle, - ConstantInt::get(I64Ty, 0), "needle_vec"); + LoadNeedle = Builder.CreateExtractVector( + FixedVectorType::get(CharTy, VF), LoadNeedle, uint64_t(0), "needle_vec"); // (2.c) Test if there's a match. Value *MatchPred = Builder.CreateIntrinsic( diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp index 6e8e1670db952..c0e9b8dc2201f 100644 --- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp +++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp @@ -5787,8 +5787,7 @@ static Value *createInsertVector( function_ref<Value *(Value *, Value *, ArrayRef<int>)> Generator = {}) { const unsigned SubVecVF = getNumElements(V->getType()); if (Index % SubVecVF == 0) { - Vec = Builder.CreateInsertVector(Vec->getType(), Vec, V, - Builder.getInt64(Index)); + Vec = Builder.CreateInsertVector(Vec->getType(), Vec, V, Index); } else { // Create shuffle, insertvector requires that index is multiple of // the subvector length. @@ -5818,7 +5817,7 @@ static Value *createExtractVector(IRBuilderBase &Builder, Value *Vec, if (Index % SubVecVF == 0) { VectorType *SubVecTy = getWidenedType(Vec->getType()->getScalarType(), SubVecVF); - return Builder.CreateExtractVector(SubVecTy, Vec, Builder.getInt64(Index)); + return Builder.CreateExtractVector(SubVecTy, Vec, Index); } // Create shuffle, extract_subvector requires that index is multiple of // the subvector length. _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
