diff options
Diffstat (limited to 'llvm/lib')
3 files changed, 64 insertions, 4 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp index fd59c3a7c0c..74f1e695ff6 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp @@ -706,12 +706,35 @@ Instruction *InstCombiner::visitTrunc(TruncInst &CI) { if (SimplifyDemandedInstructionBits(CI)) return &CI; - // Canonicalize trunc x to i1 -> (icmp ne (and x, 1), 0), likewise for vector. if (DestTy->getScalarSizeInBits() == 1) { - Constant *One = ConstantInt::get(SrcTy, 1); - Src = Builder.CreateAnd(Src, One); Value *Zero = Constant::getNullValue(Src->getType()); - return new ICmpInst(ICmpInst::ICMP_NE, Src, Zero); + if (DestTy->isIntegerTy()) { + // Canonicalize trunc x to i1 -> icmp ne (and x, 1), 0 (scalar only). + // TODO: We canonicalize to more instructions here because we are probably + // lacking equivalent analysis for trunc relative to icmp. There may also + // be codegen concerns. If those trunc limitations were removed, we could + // remove this transform. + Value *And = Builder.CreateAnd(Src, ConstantInt::get(SrcTy, 1)); + return new ICmpInst(ICmpInst::ICMP_NE, And, Zero); + } + + // For vectors, we do not canonicalize all truncs to icmp, so optimize + // patterns that would be covered within visitICmpInst. + Value *X; + const APInt *C; + if (match(Src, m_OneUse(m_LShr(m_Value(X), m_APInt(C))))) { + // trunc (lshr X, C) to i1 --> icmp ne (and X, C'), 0 + APInt MaskC = APInt(SrcTy->getScalarSizeInBits(), 1).shl(*C); + Value *And = Builder.CreateAnd(X, ConstantInt::get(SrcTy, MaskC)); + return new ICmpInst(ICmpInst::ICMP_NE, And, Zero); + } + if (match(Src, m_OneUse(m_c_Or(m_LShr(m_Value(X), m_APInt(C)), + m_Deferred(X))))) { + // trunc (or (lshr X, C), X) to i1 --> icmp ne (and X, C'), 0 + APInt MaskC = APInt(SrcTy->getScalarSizeInBits(), 1).shl(*C) | 1; + Value *And = Builder.CreateAnd(X, ConstantInt::get(SrcTy, MaskC)); + return new ICmpInst(ICmpInst::ICMP_NE, And, Zero); + } } // FIXME: Maybe combine the next two transforms to handle the no cast case diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp index 07bd98b30ab..bf8bc8818f7 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp @@ -1609,6 +1609,13 @@ Instruction *InstCombiner::foldICmpAndShift(ICmpInst &Cmp, BinaryOperator *And, Instruction *InstCombiner::foldICmpAndConstConst(ICmpInst &Cmp, BinaryOperator *And, const APInt &C1) { + // For vectors: icmp ne (and X, 1), 0 --> trunc X to N x i1 + // TODO: We canonicalize to the longer form for scalars because we have + // better analysis/folds for icmp, and codegen may be better with icmp. + if (Cmp.getPredicate() == CmpInst::ICMP_NE && Cmp.getType()->isVectorTy() && + C1.isNullValue() && match(And->getOperand(1), m_One())) + return new TruncInst(And->getOperand(0), Cmp.getType()); + const APInt *C2; if (!match(And->getOperand(1), m_APInt(C2))) return nullptr; diff --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp index 61a3e31f960..7258127f319 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp @@ -1477,6 +1477,33 @@ static Instruction *narrowVectorSelect(ShuffleVectorInst &Shuf, return SelectInst::Create(NarrowCond, NarrowX, NarrowY); } +/// Try to combine 2 shuffles into 1 shuffle by concatenating a shuffle mask. +static Instruction *foldIdentityExtractShuffle(ShuffleVectorInst &Shuf) { + Value *Op0 = Shuf.getOperand(0), *Op1 = Shuf.getOperand(1); + if (!Shuf.isIdentityWithExtract() || !isa<UndefValue>(Op1)) + return nullptr; + + Value *X, *Y; + Constant *Mask; + if (!match(Op0, m_ShuffleVector(m_Value(X), m_Value(Y), m_Constant(Mask)))) + return nullptr; + + // We are extracting a subvector from a shuffle. Remove excess elements from + // the 1st shuffle mask to eliminate the extract. + // shuf (shuf X, Y, <C0, C1, C2, C3>), undef, <0, undef, 2> --> + // shuf X, Y, <C0, undef, C2> + unsigned NumElts = Shuf.getType()->getVectorNumElements(); + SmallVector<Constant *, 16> NewMask(NumElts); + for (unsigned i = 0; i != NumElts; ++i) { + // If the extracting shuffle has an undef mask element, it transfers to the + // new shuffle mask. Otherwise, copy the original mask element. + Constant *ExtractMaskElt = Shuf.getMask()->getAggregateElement(i); + Constant *MaskElt = Mask->getAggregateElement(i); + NewMask[i] = isa<UndefValue>(ExtractMaskElt) ? ExtractMaskElt : MaskElt; + } + return new ShuffleVectorInst(X, Y, ConstantVector::get(NewMask)); +} + Instruction *InstCombiner::visitShuffleVectorInst(ShuffleVectorInst &SVI) { Value *LHS = SVI.getOperand(0); Value *RHS = SVI.getOperand(1); @@ -1499,6 +1526,9 @@ Instruction *InstCombiner::visitShuffleVectorInst(ShuffleVectorInst &SVI) { return &SVI; } + if (Instruction *I = foldIdentityExtractShuffle(SVI)) + return I; + SmallVector<int, 16> Mask = SVI.getShuffleMask(); Type *Int32Ty = Type::getInt32Ty(SVI.getContext()); unsigned LHSWidth = LHS->getType()->getVectorNumElements(); |
