diff --git a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h --- a/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h +++ b/llvm/include/llvm/Transforms/InstCombine/InstCombiner.h @@ -293,8 +293,7 @@ static Constant * getSafeVectorConstantForBinop(BinaryOperator::BinaryOps Opcode, Constant *In, bool IsRHSConstant) { - auto *InVTy = dyn_cast(In->getType()); - assert(InVTy && "Not expecting scalars here"); + auto *InVTy = cast(In->getType()); Type *EltTy = InVTy->getElementType(); auto *SafeC = ConstantExpr::getBinOpIdentity(Opcode, EltTy, IsRHSConstant); diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp @@ -1706,7 +1706,7 @@ if (C->getType()->isVectorTy()) { // Check each element of a constant vector. - unsigned NumElts = cast(C->getType())->getNumElements(); + unsigned NumElts = cast(C->getType())->getNumElements(); for (unsigned i = 0; i != NumElts; ++i) { Constant *Elt = C->getAggregateElement(i); if (!Elt) @@ -2199,7 +2199,7 @@ /// If all elements of two constant vectors are 0/-1 and inverses, return true. static bool areInverseVectorBitmasks(Constant *C1, Constant *C2) { - unsigned NumElts = cast(C1->getType())->getNumElements(); + unsigned NumElts = cast(C1->getType())->getNumElements(); for (unsigned i = 0; i != NumElts; ++i) { Constant *EltC1 = C1->getAggregateElement(i); Constant *EltC2 = C2->getAggregateElement(i); diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp @@ -538,7 +538,7 @@ if (!C) return nullptr; - auto *VecTy = cast(II.getType()); + auto *VecTy = cast(II.getType()); unsigned NumElts = VecTy->getNumElements(); // Only perform this transformation for <8 x i8> vector types. diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp @@ -875,7 +875,7 @@ // extractelement <8 x i32> (bitcast <4 x i64> %X to <8 x i32>), i32 0 Value *VecOp; if (match(Src, m_OneUse(m_ExtractElt(m_Value(VecOp), m_ConstantInt(Cst))))) { - auto *VecOpTy = cast(VecOp->getType()); + auto *VecOpTy = cast(VecOp->getType()); unsigned VecNumElts = VecOpTy->getNumElements(); // A badly fit destination size would result in an invalid cast. @@ -1538,7 +1538,7 @@ Type *MinType = nullptr; - unsigned NumElts = CVVTy->getNumElements(); + unsigned NumElts = cast(CVVTy)->getNumElements(); for (unsigned i = 0; i != NumElts; ++i) { auto *CFP = dyn_cast_or_null(CV->getAggregateElement(i)); if (!CFP) @@ -1942,7 +1942,8 @@ if (auto *VecTy = dyn_cast(Ty)) { // Handle vectors of pointers. // FIXME: what should happen for scalable vectors? - IntPtrTy = FixedVectorType::get(IntPtrTy, VecTy->getNumElements()); + IntPtrTy = FixedVectorType::get( + IntPtrTy, cast(VecTy)->getNumElements()); } Value *P = Builder.CreatePtrToInt(SrcOp, IntPtrTy); @@ -1997,13 +1998,14 @@ return nullptr; SrcTy = - FixedVectorType::get(DestTy->getElementType(), SrcTy->getNumElements()); + FixedVectorType::get(DestTy->getElementType(), + cast(SrcTy)->getNumElements()); InVal = IC.Builder.CreateBitCast(InVal, SrcTy); } bool IsBigEndian = IC.getDataLayout().isBigEndian(); - unsigned SrcElts = SrcTy->getNumElements(); - unsigned DestElts = DestTy->getNumElements(); + unsigned SrcElts = cast(SrcTy)->getNumElements(); + unsigned DestElts = cast(DestTy)->getNumElements(); assert(SrcElts != DestElts && "Element counts should be different."); @@ -2182,7 +2184,7 @@ /// Into two insertelements that do "buildvector{%inc, %inc5}". static Value *optimizeIntegerToVectorInsertions(BitCastInst &CI, InstCombinerImpl &IC) { - VectorType *DestVecTy = cast(CI.getType()); + auto *DestVecTy = cast(CI.getType()); Value *IntInput = CI.getOperand(0); SmallVector Elements(DestVecTy->getNumElements()); @@ -2222,7 +2224,8 @@ if (!VectorType::isValidElementType(DestType)) return nullptr; - unsigned NumElts = ExtElt->getVectorOperandType()->getNumElements(); + unsigned NumElts = + cast(ExtElt->getVectorOperandType())->getNumElements(); auto *NewVecType = FixedVectorType::get(DestType, NumElts); auto *NewBC = IC.Builder.CreateBitCast(ExtElt->getVectorOperand(), NewVecType, "bc"); @@ -2289,7 +2292,8 @@ if (auto *CondVTy = dyn_cast(CondTy)) { if (!DestTy->isVectorTy()) return nullptr; - if (cast(DestTy)->getNumElements() != CondVTy->getNumElements()) + if (cast(DestTy)->getNumElements() != + cast(CondVTy)->getNumElements()) return nullptr; } @@ -2625,11 +2629,12 @@ // a bitcast to a vector with the same # elts. Value *ShufOp0 = Shuf->getOperand(0); Value *ShufOp1 = Shuf->getOperand(1); - unsigned NumShufElts = Shuf->getType()->getNumElements(); + unsigned NumShufElts = + cast(Shuf->getType())->getNumElements(); unsigned NumSrcVecElts = - cast(ShufOp0->getType())->getNumElements(); + cast(ShufOp0->getType())->getNumElements(); if (Shuf->hasOneUse() && DestTy->isVectorTy() && - cast(DestTy)->getNumElements() == NumShufElts && + cast(DestTy)->getNumElements() == NumShufElts && NumShufElts == NumSrcVecElts) { BitCastInst *Tmp; // If either of the operands is a cast from CI.getType(), then @@ -2697,7 +2702,8 @@ if (VectorType *VT = dyn_cast(CI.getType())) { // Handle vectors of pointers. // FIXME: what should happen for scalable vectors? - MidTy = FixedVectorType::get(MidTy, VT->getNumElements()); + MidTy = FixedVectorType::get(MidTy, + cast(VT)->getNumElements()); } Value *NewBitCast = Builder.CreateBitCast(Src, MidTy); diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp @@ -898,7 +898,7 @@ // For vectors, we apply the same reasoning on a per-lane basis. auto *Base = GEPLHS->getPointerOperand(); if (GEPLHS->getType()->isVectorTy() && Base->getType()->isPointerTy()) { - int NumElts = cast(GEPLHS->getType())->getNumElements(); + int NumElts = cast(GEPLHS->getType())->getNumElements(); Base = Builder.CreateVectorSplat(NumElts, Base); } return new ICmpInst(Cond, Base, @@ -1886,7 +1886,8 @@ if (ExactLogBase2 != -1 && DL.isLegalInteger(ExactLogBase2 + 1)) { Type *NTy = IntegerType::get(Cmp.getContext(), ExactLogBase2 + 1); if (auto *AndVTy = dyn_cast(And->getType())) - NTy = FixedVectorType::get(NTy, AndVTy->getNumElements()); + NTy = FixedVectorType::get( + NTy, cast(AndVTy)->getNumElements()); Value *Trunc = Builder.CreateTrunc(X, NTy); auto NewPred = Cmp.getPredicate() == CmpInst::ICMP_EQ ? CmpInst::ICMP_SGE : CmpInst::ICMP_SLT; @@ -2192,7 +2193,8 @@ DL.isLegalInteger(TypeBits - Amt)) { Type *TruncTy = IntegerType::get(Cmp.getContext(), TypeBits - Amt); if (auto *ShVTy = dyn_cast(ShType)) - TruncTy = FixedVectorType::get(TruncTy, ShVTy->getNumElements()); + TruncTy = FixedVectorType::get( + TruncTy, cast(ShVTy)->getNumElements()); Constant *NewC = ConstantInt::get(TruncTy, C.ashr(*ShiftAmt).trunc(TypeBits - Amt)); return new ICmpInst(Pred, Builder.CreateTrunc(X, TruncTy), NewC); @@ -2826,7 +2828,8 @@ Type *NewType = Builder.getIntNTy(XType->getScalarSizeInBits()); if (auto *XVTy = dyn_cast(XType)) - NewType = FixedVectorType::get(NewType, XVTy->getNumElements()); + NewType = FixedVectorType::get( + NewType, cast(XVTy)->getNumElements()); Value *NewBitcast = Builder.CreateBitCast(X, NewType); if (TrueIfSigned) return new ICmpInst(ICmpInst::ICMP_SLT, NewBitcast, @@ -3391,7 +3394,7 @@ Type *OpTy = M->getType(); auto *VecC = dyn_cast(M); if (OpTy->isVectorTy() && VecC && VecC->containsUndefElement()) { - auto *OpVTy = cast(OpTy); + auto *OpVTy = cast(OpTy); Constant *SafeReplacementConstant = nullptr; for (unsigned i = 0, e = OpVTy->getNumElements(); i != e; ++i) { if (!isa(VecC->getAggregateElement(i))) { @@ -5241,7 +5244,7 @@ if (!ConstantIsOk(CI)) return llvm::None; } else if (auto *VTy = dyn_cast(Type)) { - unsigned NumElts = VTy->getNumElements(); + unsigned NumElts = cast(VTy)->getNumElements(); for (unsigned i = 0; i != NumElts; ++i) { Constant *Elt = C->getAggregateElement(i); if (!Elt) diff --git a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp @@ -1065,11 +1065,11 @@ return nullptr; } if (auto *AT = dyn_cast(VT)) { - if (AT->getNumElements() != UT->getNumElements()) + if (AT->getNumElements() != cast(UT)->getNumElements()) return nullptr; } else { auto *ST = cast(VT); - if (ST->getNumElements() != UT->getNumElements()) + if (ST->getNumElements() != cast(UT)->getNumElements()) return nullptr; for (const auto *EltT : ST->elements()) { if (EltT != UT->getElementType()) diff --git a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp @@ -1516,7 +1516,7 @@ // If it's a constant vector, flip any negative values positive. if (isa(Op1) || isa(Op1)) { Constant *C = cast(Op1); - unsigned VWidth = cast(C->getType())->getNumElements(); + unsigned VWidth = cast(C->getType())->getNumElements(); bool hasNegative = false; bool hasMissing = false; diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp @@ -305,8 +305,8 @@ if (auto *CondVTy = dyn_cast(CondTy)) { if (!FIOpndTy->isVectorTy()) return nullptr; - if (CondVTy->getNumElements() != - cast(FIOpndTy)->getNumElements()) + if (cast(CondVTy)->getNumElements() != + cast(FIOpndTy)->getNumElements()) return nullptr; // TODO: If the backend knew how to deal with casts better, we could @@ -1971,7 +1971,8 @@ if (!CondVal->getType()->isVectorTy() || !match(CondVal, m_Constant(CondC))) return nullptr; - unsigned NumElts = cast(CondVal->getType())->getNumElements(); + unsigned NumElts = + cast(CondVal->getType())->getNumElements(); SmallVector Mask; Mask.reserve(NumElts); for (unsigned i = 0; i != NumElts; ++i) { diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp @@ -392,7 +392,8 @@ if (VectorType *DstVTy = dyn_cast(I->getType())) { if (VectorType *SrcVTy = dyn_cast(I->getOperand(0)->getType())) { - if (DstVTy->getNumElements() != SrcVTy->getNumElements()) + if (cast(DstVTy)->getNumElements() != + cast(SrcVTy)->getNumElements()) // Don't touch a bitcast between vectors of different element counts. return nullptr; } else @@ -1186,8 +1187,8 @@ assert(Shuffle->getOperand(0)->getType() == Shuffle->getOperand(1)->getType() && "Expected shuffle operands to have same type"); - unsigned OpWidth = - cast(Shuffle->getOperand(0)->getType())->getNumElements(); + unsigned OpWidth = cast(Shuffle->getOperand(0)->getType()) + ->getNumElements(); // Handle trivial case of a splat. Only check the first element of LHS // operand. if (all_of(Shuffle->getShuffleMask(), [](int Elt) { return Elt == 0; }) && @@ -1288,7 +1289,8 @@ // this constant vector to single insertelement instruction. // shufflevector V, C, -> // insertelement V, C[ci], ci-n - if (OpWidth == Shuffle->getType()->getNumElements()) { + if (OpWidth == + cast(Shuffle->getType())->getNumElements()) { Value *Op = nullptr; Constant *Value = nullptr; unsigned Idx = -1u; @@ -1375,7 +1377,7 @@ // Vector->vector casts only. VectorType *VTy = dyn_cast(I->getOperand(0)->getType()); if (!VTy) break; - unsigned InVWidth = VTy->getNumElements(); + unsigned InVWidth = cast(VTy)->getNumElements(); APInt InputDemandedElts(InVWidth, 0); UndefElts2 = APInt(InVWidth, 0); unsigned Ratio; diff --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp @@ -185,15 +185,19 @@ // extelt (bitcast VecX), IndexC --> bitcast X[IndexC] auto *SrcTy = cast(X->getType()); Type *DestTy = Ext.getType(); - unsigned NumSrcElts = SrcTy->getNumElements(); - unsigned NumElts = Ext.getVectorOperandType()->getNumElements(); + ElementCount NumSrcElts = SrcTy->getElementCount(); + ElementCount NumElts = + cast(Ext.getVectorOperandType())->getElementCount(); if (NumSrcElts == NumElts) if (Value *Elt = findScalarElement(X, ExtIndexC)) return new BitCastInst(Elt, DestTy); + assert(NumSrcElts.isScalable() == NumElts.isScalable() && + "Src and Dst must be the same sort of vector type"); + // If the source elements are wider than the destination, try to shift and // truncate a subset of scalar bits of an insert op. - if (NumSrcElts < NumElts) { + if (NumSrcElts.getKnownMinValue() < NumElts.getKnownMinValue()) { Value *Scalar; uint64_t InsIndexC; if (!match(X, m_InsertElt(m_Value(), m_Value(Scalar), @@ -204,7 +208,8 @@ // into. Example: if we inserted element 1 of a <2 x i64> and we are // extracting an i16 (narrowing ratio = 4), then this extract must be from 1 // of elements 4-7 of the bitcasted vector. - unsigned NarrowingRatio = NumElts / NumSrcElts; + unsigned NarrowingRatio = + NumElts.getKnownMinValue() / NumSrcElts.getKnownMinValue(); if (ExtIndexC / NarrowingRatio != InsIndexC) return nullptr; @@ -266,7 +271,7 @@ /// Find elements of V demanded by UserInstr. static APInt findDemandedEltsBySingleUser(Value *V, Instruction *UserInstr) { - unsigned VWidth = cast(V->getType())->getNumElements(); + unsigned VWidth = cast(V->getType())->getNumElements(); // Conservatively assume that all elements are needed. APInt UsedElts(APInt::getAllOnesValue(VWidth)); @@ -284,7 +289,7 @@ case Instruction::ShuffleVector: { ShuffleVectorInst *Shuffle = cast(UserInstr); unsigned MaskNumElts = - cast(UserInstr->getType())->getNumElements(); + cast(UserInstr->getType())->getNumElements(); UsedElts = APInt(VWidth, 0); for (unsigned i = 0; i < MaskNumElts; i++) { @@ -310,7 +315,7 @@ /// no user demands an element of V, then the corresponding bit /// remains unset in the returned value. static APInt findDemandedEltsByAllUsers(Value *V) { - unsigned VWidth = cast(V->getType())->getNumElements(); + unsigned VWidth = cast(V->getType())->getNumElements(); APInt UnionUsedElts(VWidth, 0); for (const Use &U : V->uses()) { @@ -467,7 +472,7 @@ SmallVectorImpl &Mask) { assert(LHS->getType() == RHS->getType() && "Invalid CollectSingleShuffleElements"); - unsigned NumElts = cast(V->getType())->getNumElements(); + unsigned NumElts = cast(V->getType())->getNumElements(); if (isa(V)) { Mask.assign(NumElts, -1); @@ -509,7 +514,7 @@ unsigned ExtractedIdx = cast(EI->getOperand(1))->getZExtValue(); unsigned NumLHSElts = - cast(LHS->getType())->getNumElements(); + cast(LHS->getType())->getNumElements(); // This must be extracting from either LHS or RHS. if (EI->getOperand(0) == LHS || EI->getOperand(0) == RHS) { @@ -539,8 +544,8 @@ static void replaceExtractElements(InsertElementInst *InsElt, ExtractElementInst *ExtElt, InstCombinerImpl &IC) { - VectorType *InsVecType = InsElt->getType(); - VectorType *ExtVecType = ExtElt->getVectorOperandType(); + auto *InsVecType = cast(InsElt->getType()); + auto *ExtVecType = cast(ExtElt->getVectorOperandType()); unsigned NumInsElts = InsVecType->getNumElements(); unsigned NumExtElts = ExtVecType->getNumElements(); @@ -668,7 +673,7 @@ } unsigned NumLHSElts = - cast(RHS->getType())->getNumElements(); + cast(RHS->getType())->getNumElements(); Mask[InsertedIdx % NumElts] = NumLHSElts + ExtractedIdx; return std::make_pair(LR.first, RHS); } @@ -677,7 +682,8 @@ // We've gone as far as we can: anything on the other side of the // extractelement will already have been converted into a shuffle. unsigned NumLHSElts = - cast(EI->getOperand(0)->getType())->getNumElements(); + cast(EI->getOperand(0)->getType()) + ->getNumElements(); for (unsigned i = 0; i != NumElts; ++i) Mask.push_back(i == InsertedIdx ? ExtractedIdx : NumLHSElts + i); return std::make_pair(EI->getOperand(0), PermittedRHS); @@ -1144,7 +1150,8 @@ // For example: // inselt (shuf (inselt undef, X, 0), undef, <0,undef,0,undef>), X, 1 // --> shuf (inselt undef, X, 0), undef, <0,0,0,undef> - unsigned NumMaskElts = Shuf->getType()->getNumElements(); + unsigned NumMaskElts = + cast(Shuf->getType())->getNumElements(); SmallVector NewMask(NumMaskElts); for (unsigned i = 0; i != NumMaskElts; ++i) NewMask[i] = i == IdxC ? 0 : Shuf->getMaskValue(i); @@ -1182,7 +1189,8 @@ // that same index value. // For example: // inselt (shuf X, IdMask), (extelt X, IdxC), IdxC --> shuf X, IdMask' - unsigned NumMaskElts = Shuf->getType()->getNumElements(); + unsigned NumMaskElts = + cast(Shuf->getType())->getNumElements(); SmallVector NewMask(NumMaskElts); ArrayRef OldMask = Shuf->getShuffleMask(); for (unsigned i = 0; i != NumMaskElts; ++i) { @@ -1512,7 +1520,7 @@ // longer vector ops, but that may result in more expensive codegen. Type *ITy = I->getType(); if (ITy->isVectorTy() && - Mask.size() > cast(ITy)->getNumElements()) + Mask.size() > cast(ITy)->getNumElements()) return false; for (Value *Operand : I->operands()) { if (!canEvaluateShuffled(Operand, Mask, Depth - 1)) @@ -1670,7 +1678,8 @@ case Instruction::GetElementPtr: { SmallVector NewOps; bool NeedsRebuild = - (Mask.size() != cast(I->getType())->getNumElements()); + (Mask.size() != + cast(I->getType())->getNumElements()); for (int i = 0, e = I->getNumOperands(); i != e; ++i) { Value *V; // Recursively call evaluateInDifferentElementOrder on vector arguments @@ -1725,7 +1734,7 @@ static bool isShuffleExtractingFromLHS(ShuffleVectorInst &SVI, ArrayRef Mask) { unsigned LHSElems = - cast(SVI.getOperand(0)->getType())->getNumElements(); + cast(SVI.getOperand(0)->getType())->getNumElements(); unsigned MaskElems = Mask.size(); unsigned BegIdx = Mask.front(); unsigned EndIdx = Mask.back(); @@ -1857,7 +1866,8 @@ // For example: // shuf (inselt undef, X, 2), undef, <2,2,undef> // --> shuf (inselt undef, X, 0), undef, <0,0,undef> - unsigned NumMaskElts = Shuf.getType()->getNumElements(); + unsigned NumMaskElts = + cast(Shuf.getType())->getNumElements(); SmallVector NewMask(NumMaskElts, 0); for (unsigned i = 0; i != NumMaskElts; ++i) if (Mask[i] == UndefMaskElem) @@ -1875,7 +1885,7 @@ // Canonicalize to choose from operand 0 first unless operand 1 is undefined. // Commuting undef to operand 0 conflicts with another canonicalization. - unsigned NumElts = Shuf.getType()->getNumElements(); + unsigned NumElts = cast(Shuf.getType())->getNumElements(); if (!isa(Shuf.getOperand(1)) && Shuf.getMaskValue(0) >= (int)NumElts) { // TODO: Can we assert that both operands of a shuffle-select are not undef @@ -2010,8 +2020,8 @@ // and the source element type must be larger than the shuffle element type. Type *SrcType = X->getType(); if (!SrcType->isVectorTy() || !SrcType->isIntOrIntVectorTy() || - cast(SrcType)->getNumElements() != - cast(DestType)->getNumElements() || + cast(SrcType)->getNumElements() != + cast(DestType)->getNumElements() || SrcType->getScalarSizeInBits() % DestType->getScalarSizeInBits() != 0) return nullptr; @@ -2055,10 +2065,11 @@ // We need a narrow condition value. It must be extended with undef elements // and have the same number of elements as this shuffle. - unsigned NarrowNumElts = Shuf.getType()->getNumElements(); + unsigned NarrowNumElts = + cast(Shuf.getType())->getNumElements(); Value *NarrowCond; if (!match(Cond, m_OneUse(m_Shuffle(m_Value(NarrowCond), m_Undef()))) || - cast(NarrowCond->getType())->getNumElements() != + cast(NarrowCond->getType())->getNumElements() != NarrowNumElts || !cast(Cond)->isIdentityWithPadding()) return nullptr; @@ -2098,7 +2109,7 @@ // new shuffle mask. Otherwise, copy the original mask element. Example: // shuf (shuf X, Y, ), undef, <0, undef, 2, 3> --> // shuf X, Y, - unsigned NumElts = Shuf.getType()->getNumElements(); + unsigned NumElts = cast(Shuf.getType())->getNumElements(); SmallVector NewMask(NumElts); assert(NumElts < Mask.size() && "Identity with extract must have less elements than its inputs"); @@ -2123,7 +2134,7 @@ // TODO: This restriction could be removed if the insert has only one use // (because the transform would require a new length-changing shuffle). int NumElts = Mask.size(); - if (NumElts != (int)(cast(V0->getType())->getNumElements())) + if (NumElts != (int)(cast(V0->getType())->getNumElements())) return nullptr; // This is a specialization of a fold in SimplifyDemandedVectorElts. We may @@ -2218,9 +2229,10 @@ Value *X = Shuffle0->getOperand(0); Value *Y = Shuffle1->getOperand(0); if (X->getType() != Y->getType() || - !isPowerOf2_32(Shuf.getType()->getNumElements()) || - !isPowerOf2_32(Shuffle0->getType()->getNumElements()) || - !isPowerOf2_32(cast(X->getType())->getNumElements()) || + !isPowerOf2_32(cast(Shuf.getType())->getNumElements()) || + !isPowerOf2_32( + cast(Shuffle0->getType())->getNumElements()) || + !isPowerOf2_32(cast(X->getType())->getNumElements()) || isa(X) || isa(Y)) return nullptr; assert(isa(Shuffle0->getOperand(1)) && @@ -2231,8 +2243,8 @@ // operands directly by adjusting the shuffle mask to account for the narrower // types: // shuf (widen X), (widen Y), Mask --> shuf X, Y, Mask' - int NarrowElts = cast(X->getType())->getNumElements(); - int WideElts = Shuffle0->getType()->getNumElements(); + int NarrowElts = cast(X->getType())->getNumElements(); + int WideElts = cast(Shuffle0->getType())->getNumElements(); assert(WideElts > NarrowElts && "Unexpected types for identity with padding"); ArrayRef Mask = Shuf.getShuffleMask(); @@ -2273,9 +2285,13 @@ SVI.getType(), ShufQuery)) return replaceInstUsesWith(SVI, V); + // Bail out for scalable vectors + if (isa(LHS->getType())) + return nullptr; + // shuffle x, x, mask --> shuffle x, undef, mask' - unsigned VWidth = SVI.getType()->getNumElements(); - unsigned LHSWidth = cast(LHS->getType())->getNumElements(); + unsigned VWidth = cast(SVI.getType())->getNumElements(); + unsigned LHSWidth = cast(LHS->getType())->getNumElements(); ArrayRef Mask = SVI.getShuffleMask(); Type *Int32Ty = Type::getInt32Ty(SVI.getContext()); @@ -2289,7 +2305,7 @@ if (match(LHS, m_BitCast(m_Value(X))) && match(RHS, m_Undef()) && X->getType()->isVectorTy() && VWidth == LHSWidth) { // Try to create a scaled mask constant. - auto *XType = cast(X->getType()); + auto *XType = cast(X->getType()); unsigned XNumElts = XType->getNumElements(); SmallVector ScaledMask; if (XNumElts >= VWidth) { @@ -2397,7 +2413,7 @@ if (isShuffleExtractingFromLHS(SVI, Mask)) { Value *V = LHS; unsigned MaskElems = Mask.size(); - VectorType *SrcTy = cast(V->getType()); + auto *SrcTy = cast(V->getType()); unsigned VecBitWidth = SrcTy->getPrimitiveSizeInBits().getFixedSize(); unsigned SrcElemBitWidth = DL.getTypeSizeInBits(SrcTy->getElementType()); assert(SrcElemBitWidth && "vector elements must have a bitwidth"); @@ -2515,11 +2531,11 @@ if (LHSShuffle) { LHSOp0 = LHSShuffle->getOperand(0); LHSOp1 = LHSShuffle->getOperand(1); - LHSOp0Width = cast(LHSOp0->getType())->getNumElements(); + LHSOp0Width = cast(LHSOp0->getType())->getNumElements(); } if (RHSShuffle) { RHSOp0 = RHSShuffle->getOperand(0); - RHSOp0Width = cast(RHSOp0->getType())->getNumElements(); + RHSOp0Width = cast(RHSOp0->getType())->getNumElements(); } Value* newLHS = LHS; Value* newRHS = RHS; diff --git a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp --- a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp +++ b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp @@ -959,7 +959,8 @@ return nullptr; // If vectors, verify that they have the same number of elements. - if (SrcTy && SrcTy->getNumElements() != DestTy->getNumElements()) + if (SrcTy && cast(SrcTy)->getNumElements() != + cast(DestTy)->getNumElements()) return nullptr; } @@ -2371,7 +2372,7 @@ // gep (bitcast [c x ty]* X to *), Y, Z --> gep X, Y, Z auto areMatchingArrayAndVecTypes = [](Type *ArrTy, Type *VecTy, const DataLayout &DL) { - auto *VecVTy = cast(VecTy); + auto *VecVTy = cast(VecTy); return ArrTy->getArrayElementType() == VecVTy->getElementType() && ArrTy->getArrayNumElements() == VecVTy->getNumElements() && DL.getTypeAllocSize(ArrTy) == DL.getTypeAllocSize(VecTy);