//===- InstCombineCasts.cpp -----------------------------------------------===// // // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. // See https://llvm.org/LICENSE.txt for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //===----------------------------------------------------------------------===// // // This file implements the visit functions for cast operations. // //===----------------------------------------------------------------------===// #include "InstCombineInternal.h" #include "llvm/ADT/SetVector.h" #include "llvm/Analysis/ConstantFolding.h" #include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/IR/DIBuilder.h" #include "llvm/IR/DataLayout.h" #include "llvm/IR/PatternMatch.h" #include "llvm/Support/KnownBits.h" #include "llvm/Transforms/InstCombine/InstCombiner.h" #include using namespace llvm; using namespace PatternMatch; #define DEBUG_TYPE "instcombine" /// Analyze 'Val', seeing if it is a simple linear expression. /// If so, decompose it, returning some value X, such that Val is /// X*Scale+Offset. /// static Value *decomposeSimpleLinearExpr(Value *Val, unsigned &Scale, uint64_t &Offset) { if (ConstantInt *CI = dyn_cast(Val)) { Offset = CI->getZExtValue(); Scale = 0; return ConstantInt::get(Val->getType(), 0); } if (BinaryOperator *I = dyn_cast(Val)) { // Cannot look past anything that might overflow. OverflowingBinaryOperator *OBI = dyn_cast(Val); if (OBI && !OBI->hasNoUnsignedWrap() && !OBI->hasNoSignedWrap()) { Scale = 1; Offset = 0; return Val; } if (ConstantInt *RHS = dyn_cast(I->getOperand(1))) { if (I->getOpcode() == Instruction::Shl) { // This is a value scaled by '1 << the shift amt'. Scale = UINT64_C(1) << RHS->getZExtValue(); Offset = 0; return I->getOperand(0); } if (I->getOpcode() == Instruction::Mul) { // This value is scaled by 'RHS'. Scale = RHS->getZExtValue(); Offset = 0; return I->getOperand(0); } if (I->getOpcode() == Instruction::Add) { // We have X+C. Check to see if we really have (X*C2)+C1, // where C1 is divisible by C2. unsigned SubScale; Value *SubVal = decomposeSimpleLinearExpr(I->getOperand(0), SubScale, Offset); Offset += RHS->getZExtValue(); Scale = SubScale; return SubVal; } } } // Otherwise, we can't look past this. Scale = 1; Offset = 0; return Val; } /// If we find a cast of an allocation instruction, try to eliminate the cast by /// moving the type information into the alloc. Instruction *InstCombinerImpl::PromoteCastOfAllocation(BitCastInst &CI, AllocaInst &AI) { PointerType *PTy = cast(CI.getType()); IRBuilderBase::InsertPointGuard Guard(Builder); Builder.SetInsertPoint(&AI); // Get the type really allocated and the type casted to. Type *AllocElTy = AI.getAllocatedType(); Type *CastElTy = PTy->getElementType(); if (!AllocElTy->isSized() || !CastElTy->isSized()) return nullptr; // This optimisation does not work for cases where the cast type // is scalable and the allocated type is not. This because we need to // know how many times the casted type fits into the allocated type. // For the opposite case where the allocated type is scalable and the // cast type is not this leads to poor code quality due to the // introduction of 'vscale' into the calculations. It seems better to // bail out for this case too until we've done a proper cost-benefit // analysis. bool AllocIsScalable = isa(AllocElTy); bool CastIsScalable = isa(CastElTy); if (AllocIsScalable != CastIsScalable) return nullptr; Align AllocElTyAlign = DL.getABITypeAlign(AllocElTy); Align CastElTyAlign = DL.getABITypeAlign(CastElTy); if (CastElTyAlign < AllocElTyAlign) return nullptr; // If the allocation has multiple uses, only promote it if we are strictly // increasing the alignment of the resultant allocation. If we keep it the // same, we open the door to infinite loops of various kinds. if (!AI.hasOneUse() && CastElTyAlign == AllocElTyAlign) return nullptr; // The alloc and cast types should be either both fixed or both scalable. uint64_t AllocElTySize = DL.getTypeAllocSize(AllocElTy).getKnownMinSize(); uint64_t CastElTySize = DL.getTypeAllocSize(CastElTy).getKnownMinSize(); if (CastElTySize == 0 || AllocElTySize == 0) return nullptr; // If the allocation has multiple uses, only promote it if we're not // shrinking the amount of memory being allocated. uint64_t AllocElTyStoreSize = DL.getTypeStoreSize(AllocElTy).getKnownMinSize(); uint64_t CastElTyStoreSize = DL.getTypeStoreSize(CastElTy).getKnownMinSize(); if (!AI.hasOneUse() && CastElTyStoreSize < AllocElTyStoreSize) return nullptr; // See if we can satisfy the modulus by pulling a scale out of the array // size argument. unsigned ArraySizeScale; uint64_t ArrayOffset; Value *NumElements = // See if the array size is a decomposable linear expr. decomposeSimpleLinearExpr(AI.getOperand(0), ArraySizeScale, ArrayOffset); // If we can now satisfy the modulus, by using a non-1 scale, we really can // do the xform. if ((AllocElTySize*ArraySizeScale) % CastElTySize != 0 || (AllocElTySize*ArrayOffset ) % CastElTySize != 0) return nullptr; // We don't currently support arrays of scalable types. assert(!AllocIsScalable || (ArrayOffset == 1 && ArraySizeScale == 0)); unsigned Scale = (AllocElTySize*ArraySizeScale)/CastElTySize; Value *Amt = nullptr; if (Scale == 1) { Amt = NumElements; } else { Amt = ConstantInt::get(AI.getArraySize()->getType(), Scale); // Insert before the alloca, not before the cast. Amt = Builder.CreateMul(Amt, NumElements); } if (uint64_t Offset = (AllocElTySize*ArrayOffset)/CastElTySize) { Value *Off = ConstantInt::get(AI.getArraySize()->getType(), Offset, true); Amt = Builder.CreateAdd(Amt, Off); } AllocaInst *New = Builder.CreateAlloca(CastElTy, Amt); New->setAlignment(AI.getAlign()); New->takeName(&AI); New->setUsedWithInAlloca(AI.isUsedWithInAlloca()); // If the allocation has multiple real uses, insert a cast and change all // things that used it to use the new cast. This will also hack on CI, but it // will die soon. if (!AI.hasOneUse()) { // New is the allocation instruction, pointer typed. AI is the original // allocation instruction, also pointer typed. Thus, cast to use is BitCast. Value *NewCast = Builder.CreateBitCast(New, AI.getType(), "tmpcast"); replaceInstUsesWith(AI, NewCast); eraseInstFromFunction(AI); } return replaceInstUsesWith(CI, New); } /// Given an expression that CanEvaluateTruncated or CanEvaluateSExtd returns /// true for, actually insert the code to evaluate the expression. Value *InstCombinerImpl::EvaluateInDifferentType(Value *V, Type *Ty, bool isSigned) { if (Constant *C = dyn_cast(V)) { C = ConstantExpr::getIntegerCast(C, Ty, isSigned /*Sext or ZExt*/); // If we got a constantexpr back, try to simplify it with DL info. return ConstantFoldConstant(C, DL, &TLI); } // Otherwise, it must be an instruction. Instruction *I = cast(V); Instruction *Res = nullptr; unsigned Opc = I->getOpcode(); switch (Opc) { case Instruction::Add: case Instruction::Sub: case Instruction::Mul: case Instruction::And: case Instruction::Or: case Instruction::Xor: case Instruction::AShr: case Instruction::LShr: case Instruction::Shl: case Instruction::UDiv: case Instruction::URem: { Value *LHS = EvaluateInDifferentType(I->getOperand(0), Ty, isSigned); Value *RHS = EvaluateInDifferentType(I->getOperand(1), Ty, isSigned); Res = BinaryOperator::Create((Instruction::BinaryOps)Opc, LHS, RHS); break; } case Instruction::Trunc: case Instruction::ZExt: case Instruction::SExt: // If the source type of the cast is the type we're trying for then we can // just return the source. There's no need to insert it because it is not // new. if (I->getOperand(0)->getType() == Ty) return I->getOperand(0); // Otherwise, must be the same type of cast, so just reinsert a new one. // This also handles the case of zext(trunc(x)) -> zext(x). Res = CastInst::CreateIntegerCast(I->getOperand(0), Ty, Opc == Instruction::SExt); break; case Instruction::Select: { Value *True = EvaluateInDifferentType(I->getOperand(1), Ty, isSigned); Value *False = EvaluateInDifferentType(I->getOperand(2), Ty, isSigned); Res = SelectInst::Create(I->getOperand(0), True, False); break; } case Instruction::PHI: { PHINode *OPN = cast(I); PHINode *NPN = PHINode::Create(Ty, OPN->getNumIncomingValues()); for (unsigned i = 0, e = OPN->getNumIncomingValues(); i != e; ++i) { Value *V = EvaluateInDifferentType(OPN->getIncomingValue(i), Ty, isSigned); NPN->addIncoming(V, OPN->getIncomingBlock(i)); } Res = NPN; break; } default: // TODO: Can handle more cases here. llvm_unreachable("Unreachable!"); } Res->takeName(I); return InsertNewInstWith(Res, *I); } Instruction::CastOps InstCombinerImpl::isEliminableCastPair(const CastInst *CI1, const CastInst *CI2) { Type *SrcTy = CI1->getSrcTy(); Type *MidTy = CI1->getDestTy(); Type *DstTy = CI2->getDestTy(); Instruction::CastOps firstOp = CI1->getOpcode(); Instruction::CastOps secondOp = CI2->getOpcode(); Type *SrcIntPtrTy = SrcTy->isPtrOrPtrVectorTy() ? DL.getIntPtrType(SrcTy) : nullptr; Type *MidIntPtrTy = MidTy->isPtrOrPtrVectorTy() ? DL.getIntPtrType(MidTy) : nullptr; Type *DstIntPtrTy = DstTy->isPtrOrPtrVectorTy() ? DL.getIntPtrType(DstTy) : nullptr; unsigned Res = CastInst::isEliminableCastPair(firstOp, secondOp, SrcTy, MidTy, DstTy, SrcIntPtrTy, MidIntPtrTy, DstIntPtrTy); // We don't want to form an inttoptr or ptrtoint that converts to an integer // type that differs from the pointer size. if ((Res == Instruction::IntToPtr && SrcTy != DstIntPtrTy) || (Res == Instruction::PtrToInt && DstTy != SrcIntPtrTy)) Res = 0; return Instruction::CastOps(Res); } /// Implement the transforms common to all CastInst visitors. Instruction *InstCombinerImpl::commonCastTransforms(CastInst &CI) { Value *Src = CI.getOperand(0); Type *Ty = CI.getType(); // Try to eliminate a cast of a cast. if (auto *CSrc = dyn_cast(Src)) { // A->B->C cast if (Instruction::CastOps NewOpc = isEliminableCastPair(CSrc, &CI)) { // The first cast (CSrc) is eliminable so we need to fix up or replace // the second cast (CI). CSrc will then have a good chance of being dead. auto *Res = CastInst::Create(NewOpc, CSrc->getOperand(0), Ty); // Point debug users of the dying cast to the new one. if (CSrc->hasOneUse()) replaceAllDbgUsesWith(*CSrc, *Res, CI, DT); return Res; } } if (auto *Sel = dyn_cast(Src)) { // We are casting a select. Try to fold the cast into the select if the // select does not have a compare instruction with matching operand types // or the select is likely better done in a narrow type. // Creating a select with operands that are different sizes than its // condition may inhibit other folds and lead to worse codegen. auto *Cmp = dyn_cast(Sel->getCondition()); if (!Cmp || Cmp->getOperand(0)->getType() != Sel->getType() || (CI.getOpcode() == Instruction::Trunc && shouldChangeType(CI.getSrcTy(), CI.getType()))) { if (Instruction *NV = FoldOpIntoSelect(CI, Sel)) { replaceAllDbgUsesWith(*Sel, *NV, CI, DT); return NV; } } } // If we are casting a PHI, then fold the cast into the PHI. if (auto *PN = dyn_cast(Src)) { // Don't do this if it would create a PHI node with an illegal type from a // legal type. if (!Src->getType()->isIntegerTy() || !CI.getType()->isIntegerTy() || shouldChangeType(CI.getSrcTy(), CI.getType())) if (Instruction *NV = foldOpIntoPhi(CI, PN)) return NV; } // Canonicalize a unary shuffle after the cast if neither operation changes // the size or element size of the input vector. // TODO: We could allow size-changing ops if that doesn't harm codegen. // cast (shuffle X, Mask) --> shuffle (cast X), Mask Value *X; ArrayRef Mask; if (match(Src, m_OneUse(m_Shuffle(m_Value(X), m_Undef(), m_Mask(Mask))))) { // TODO: Allow scalable vectors? auto *SrcTy = dyn_cast(X->getType()); auto *DestTy = dyn_cast(Ty); if (SrcTy && DestTy && SrcTy->getNumElements() == DestTy->getNumElements() && SrcTy->getPrimitiveSizeInBits() == DestTy->getPrimitiveSizeInBits()) { Value *CastX = Builder.CreateCast(CI.getOpcode(), X, DestTy); return new ShuffleVectorInst(CastX, UndefValue::get(DestTy), Mask); } } return nullptr; } /// Constants and extensions/truncates from the destination type are always /// free to be evaluated in that type. This is a helper for canEvaluate*. static bool canAlwaysEvaluateInType(Value *V, Type *Ty) { if (isa(V)) return true; Value *X; if ((match(V, m_ZExtOrSExt(m_Value(X))) || match(V, m_Trunc(m_Value(X)))) && X->getType() == Ty) return true; return false; } /// Filter out values that we can not evaluate in the destination type for free. /// This is a helper for canEvaluate*. static bool canNotEvaluateInType(Value *V, Type *Ty) { assert(!isa(V) && "Constant should already be handled."); if (!isa(V)) return true; // We don't extend or shrink something that has multiple uses -- doing so // would require duplicating the instruction which isn't profitable. if (!V->hasOneUse()) return true; return false; } /// Return true if we can evaluate the specified expression tree as type Ty /// instead of its larger type, and arrive with the same value. /// This is used by code that tries to eliminate truncates. /// /// Ty will always be a type smaller than V. We should return true if trunc(V) /// can be computed by computing V in the smaller type. If V is an instruction, /// then trunc(inst(x,y)) can be computed as inst(trunc(x),trunc(y)), which only /// makes sense if x and y can be efficiently truncated. /// /// This function works on both vectors and scalars. /// static bool canEvaluateTruncated(Value *V, Type *Ty, InstCombinerImpl &IC, Instruction *CxtI) { if (canAlwaysEvaluateInType(V, Ty)) return true; if (canNotEvaluateInType(V, Ty)) return false; auto *I = cast(V); Type *OrigTy = V->getType(); switch (I->getOpcode()) { case Instruction::Add: case Instruction::Sub: case Instruction::Mul: case Instruction::And: case Instruction::Or: case Instruction::Xor: // These operators can all arbitrarily be extended or truncated. return canEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI) && canEvaluateTruncated(I->getOperand(1), Ty, IC, CxtI); case Instruction::UDiv: case Instruction::URem: { // UDiv and URem can be truncated if all the truncated bits are zero. uint32_t OrigBitWidth = OrigTy->getScalarSizeInBits(); uint32_t BitWidth = Ty->getScalarSizeInBits(); assert(BitWidth < OrigBitWidth && "Unexpected bitwidths!"); APInt Mask = APInt::getBitsSetFrom(OrigBitWidth, BitWidth); if (IC.MaskedValueIsZero(I->getOperand(0), Mask, 0, CxtI) && IC.MaskedValueIsZero(I->getOperand(1), Mask, 0, CxtI)) { return canEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI) && canEvaluateTruncated(I->getOperand(1), Ty, IC, CxtI); } break; } case Instruction::Shl: { // If we are truncating the result of this SHL, and if it's a shift of an // inrange amount, we can always perform a SHL in a smaller type. uint32_t BitWidth = Ty->getScalarSizeInBits(); KnownBits AmtKnownBits = llvm::computeKnownBits(I->getOperand(1), IC.getDataLayout()); if (AmtKnownBits.getMaxValue().ult(BitWidth)) return canEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI) && canEvaluateTruncated(I->getOperand(1), Ty, IC, CxtI); break; } case Instruction::LShr: { // If this is a truncate of a logical shr, we can truncate it to a smaller // lshr iff we know that the bits we would otherwise be shifting in are // already zeros. // TODO: It is enough to check that the bits we would be shifting in are // zero - use AmtKnownBits.getMaxValue(). uint32_t OrigBitWidth = OrigTy->getScalarSizeInBits(); uint32_t BitWidth = Ty->getScalarSizeInBits(); KnownBits AmtKnownBits = llvm::computeKnownBits(I->getOperand(1), IC.getDataLayout()); APInt ShiftedBits = APInt::getBitsSetFrom(OrigBitWidth, BitWidth); if (AmtKnownBits.getMaxValue().ult(BitWidth) && IC.MaskedValueIsZero(I->getOperand(0), ShiftedBits, 0, CxtI)) { return canEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI) && canEvaluateTruncated(I->getOperand(1), Ty, IC, CxtI); } break; } case Instruction::AShr: { // If this is a truncate of an arithmetic shr, we can truncate it to a // smaller ashr iff we know that all the bits from the sign bit of the // original type and the sign bit of the truncate type are similar. // TODO: It is enough to check that the bits we would be shifting in are // similar to sign bit of the truncate type. uint32_t OrigBitWidth = OrigTy->getScalarSizeInBits(); uint32_t BitWidth = Ty->getScalarSizeInBits(); KnownBits AmtKnownBits = llvm::computeKnownBits(I->getOperand(1), IC.getDataLayout()); unsigned ShiftedBits = OrigBitWidth - BitWidth; if (AmtKnownBits.getMaxValue().ult(BitWidth) && ShiftedBits < IC.ComputeNumSignBits(I->getOperand(0), 0, CxtI)) return canEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI) && canEvaluateTruncated(I->getOperand(1), Ty, IC, CxtI); break; } case Instruction::Trunc: // trunc(trunc(x)) -> trunc(x) return true; case Instruction::ZExt: case Instruction::SExt: // trunc(ext(x)) -> ext(x) if the source type is smaller than the new dest // trunc(ext(x)) -> trunc(x) if the source type is larger than the new dest return true; case Instruction::Select: { SelectInst *SI = cast(I); return canEvaluateTruncated(SI->getTrueValue(), Ty, IC, CxtI) && canEvaluateTruncated(SI->getFalseValue(), Ty, IC, CxtI); } case Instruction::PHI: { // We can change a phi if we can change all operands. Note that we never // get into trouble with cyclic PHIs here because we only consider // instructions with a single use. PHINode *PN = cast(I); for (Value *IncValue : PN->incoming_values()) if (!canEvaluateTruncated(IncValue, Ty, IC, CxtI)) return false; return true; } default: // TODO: Can handle more cases here. break; } return false; } /// Given a vector that is bitcast to an integer, optionally logically /// right-shifted, and truncated, convert it to an extractelement. /// Example (big endian): /// trunc (lshr (bitcast <4 x i32> %X to i128), 32) to i32 /// ---> /// extractelement <4 x i32> %X, 1 static Instruction *foldVecTruncToExtElt(TruncInst &Trunc, InstCombinerImpl &IC) { Value *TruncOp = Trunc.getOperand(0); Type *DestType = Trunc.getType(); if (!TruncOp->hasOneUse() || !isa(DestType)) return nullptr; Value *VecInput = nullptr; ConstantInt *ShiftVal = nullptr; if (!match(TruncOp, m_CombineOr(m_BitCast(m_Value(VecInput)), m_LShr(m_BitCast(m_Value(VecInput)), m_ConstantInt(ShiftVal)))) || !isa(VecInput->getType())) return nullptr; VectorType *VecType = cast(VecInput->getType()); unsigned VecWidth = VecType->getPrimitiveSizeInBits(); unsigned DestWidth = DestType->getPrimitiveSizeInBits(); unsigned ShiftAmount = ShiftVal ? ShiftVal->getZExtValue() : 0; if ((VecWidth % DestWidth != 0) || (ShiftAmount % DestWidth != 0)) return nullptr; // If the element type of the vector doesn't match the result type, // bitcast it to a vector type that we can extract from. unsigned NumVecElts = VecWidth / DestWidth; if (VecType->getElementType() != DestType) { VecType = FixedVectorType::get(DestType, NumVecElts); VecInput = IC.Builder.CreateBitCast(VecInput, VecType, "bc"); } unsigned Elt = ShiftAmount / DestWidth; if (IC.getDataLayout().isBigEndian()) Elt = NumVecElts - 1 - Elt; return ExtractElementInst::Create(VecInput, IC.Builder.getInt32(Elt)); } /// Funnel/Rotate left/right may occur in a wider type than necessary because of /// type promotion rules. Try to narrow the inputs and convert to funnel shift. Instruction *InstCombinerImpl::narrowFunnelShift(TruncInst &Trunc) { assert((isa(Trunc.getSrcTy()) || shouldChangeType(Trunc.getSrcTy(), Trunc.getType())) && "Don't narrow to an illegal scalar type"); // Bail out on strange types. It is possible to handle some of these patterns // even with non-power-of-2 sizes, but it is not a likely scenario. Type *DestTy = Trunc.getType(); unsigned NarrowWidth = DestTy->getScalarSizeInBits(); unsigned WideWidth = Trunc.getSrcTy()->getScalarSizeInBits(); if (!isPowerOf2_32(NarrowWidth)) return nullptr; // First, find an or'd pair of opposite shifts: // trunc (or (lshr ShVal0, ShAmt0), (shl ShVal1, ShAmt1)) BinaryOperator *Or0, *Or1; if (!match(Trunc.getOperand(0), m_OneUse(m_Or(m_BinOp(Or0), m_BinOp(Or1))))) return nullptr; Value *ShVal0, *ShVal1, *ShAmt0, *ShAmt1; if (!match(Or0, m_OneUse(m_LogicalShift(m_Value(ShVal0), m_Value(ShAmt0)))) || !match(Or1, m_OneUse(m_LogicalShift(m_Value(ShVal1), m_Value(ShAmt1)))) || Or0->getOpcode() == Or1->getOpcode()) return nullptr; // Canonicalize to or(shl(ShVal0, ShAmt0), lshr(ShVal1, ShAmt1)). if (Or0->getOpcode() == BinaryOperator::LShr) { std::swap(Or0, Or1); std::swap(ShVal0, ShVal1); std::swap(ShAmt0, ShAmt1); } assert(Or0->getOpcode() == BinaryOperator::Shl && Or1->getOpcode() == BinaryOperator::LShr && "Illegal or(shift,shift) pair"); // Match the shift amount operands for a funnel/rotate pattern. This always // matches a subtraction on the R operand. auto matchShiftAmount = [&](Value *L, Value *R, unsigned Width) -> Value * { // The shift amounts may add up to the narrow bit width: // (shl ShVal0, L) | (lshr ShVal1, Width - L) // If this is a funnel shift (different operands are shifted), then the // shift amount can not over-shift (create poison) in the narrow type. unsigned MaxShiftAmountWidth = Log2_32(NarrowWidth); APInt HiBitMask = ~APInt::getLowBitsSet(WideWidth, MaxShiftAmountWidth); if (ShVal0 == ShVal1 || MaskedValueIsZero(L, HiBitMask)) if (match(R, m_OneUse(m_Sub(m_SpecificInt(Width), m_Specific(L))))) return L; // The following patterns currently only work for rotation patterns. // TODO: Add more general funnel-shift compatible patterns. if (ShVal0 != ShVal1) return nullptr; // The shift amount may be masked with negation: // (shl ShVal0, (X & (Width - 1))) | (lshr ShVal1, ((-X) & (Width - 1))) Value *X; unsigned Mask = Width - 1; if (match(L, m_And(m_Value(X), m_SpecificInt(Mask))) && match(R, m_And(m_Neg(m_Specific(X)), m_SpecificInt(Mask)))) return X; // Same as above, but the shift amount may be extended after masking: if (match(L, m_ZExt(m_And(m_Value(X), m_SpecificInt(Mask)))) && match(R, m_ZExt(m_And(m_Neg(m_Specific(X)), m_SpecificInt(Mask))))) return X; return nullptr; }; Value *ShAmt = matchShiftAmount(ShAmt0, ShAmt1, NarrowWidth); bool IsFshl = true; // Sub on LSHR. if (!ShAmt) { ShAmt = matchShiftAmount(ShAmt1, ShAmt0, NarrowWidth); IsFshl = false; // Sub on SHL. } if (!ShAmt) return nullptr; // The right-shifted value must have high zeros in the wide type (for example // from 'zext', 'and' or 'shift'). High bits of the left-shifted value are // truncated, so those do not matter. APInt HiBitMask = APInt::getHighBitsSet(WideWidth, WideWidth - NarrowWidth); if (!MaskedValueIsZero(ShVal1, HiBitMask, 0, &Trunc)) return nullptr; // We have an unnecessarily wide rotate! // trunc (or (shl ShVal0, ShAmt), (lshr ShVal1, BitWidth - ShAmt)) // Narrow the inputs and convert to funnel shift intrinsic: // llvm.fshl.i8(trunc(ShVal), trunc(ShVal), trunc(ShAmt)) Value *NarrowShAmt = Builder.CreateTrunc(ShAmt, DestTy); Value *X, *Y; X = Y = Builder.CreateTrunc(ShVal0, DestTy); if (ShVal0 != ShVal1) Y = Builder.CreateTrunc(ShVal1, DestTy); Intrinsic::ID IID = IsFshl ? Intrinsic::fshl : Intrinsic::fshr; Function *F = Intrinsic::getDeclaration(Trunc.getModule(), IID, DestTy); return CallInst::Create(F, {X, Y, NarrowShAmt}); } /// Try to narrow the width of math or bitwise logic instructions by pulling a /// truncate ahead of binary operators. /// TODO: Transforms for truncated shifts should be moved into here. Instruction *InstCombinerImpl::narrowBinOp(TruncInst &Trunc) { Type *SrcTy = Trunc.getSrcTy(); Type *DestTy = Trunc.getType(); if (!isa(SrcTy) && !shouldChangeType(SrcTy, DestTy)) return nullptr; BinaryOperator *BinOp; if (!match(Trunc.getOperand(0), m_OneUse(m_BinOp(BinOp)))) return nullptr; Value *BinOp0 = BinOp->getOperand(0); Value *BinOp1 = BinOp->getOperand(1); switch (BinOp->getOpcode()) { case Instruction::And: case Instruction::Or: case Instruction::Xor: case Instruction::Add: case Instruction::Sub: case Instruction::Mul: { Constant *C; if (match(BinOp0, m_Constant(C))) { // trunc (binop C, X) --> binop (trunc C', X) Constant *NarrowC = ConstantExpr::getTrunc(C, DestTy); Value *TruncX = Builder.CreateTrunc(BinOp1, DestTy); return BinaryOperator::Create(BinOp->getOpcode(), NarrowC, TruncX); } if (match(BinOp1, m_Constant(C))) { // trunc (binop X, C) --> binop (trunc X, C') Constant *NarrowC = ConstantExpr::getTrunc(C, DestTy); Value *TruncX = Builder.CreateTrunc(BinOp0, DestTy); return BinaryOperator::Create(BinOp->getOpcode(), TruncX, NarrowC); } Value *X; if (match(BinOp0, m_ZExtOrSExt(m_Value(X))) && X->getType() == DestTy) { // trunc (binop (ext X), Y) --> binop X, (trunc Y) Value *NarrowOp1 = Builder.CreateTrunc(BinOp1, DestTy); return BinaryOperator::Create(BinOp->getOpcode(), X, NarrowOp1); } if (match(BinOp1, m_ZExtOrSExt(m_Value(X))) && X->getType() == DestTy) { // trunc (binop Y, (ext X)) --> binop (trunc Y), X Value *NarrowOp0 = Builder.CreateTrunc(BinOp0, DestTy); return BinaryOperator::Create(BinOp->getOpcode(), NarrowOp0, X); } break; } default: break; } if (Instruction *NarrowOr = narrowFunnelShift(Trunc)) return NarrowOr; return nullptr; } /// Try to narrow the width of a splat shuffle. This could be generalized to any /// shuffle with a constant operand, but we limit the transform to avoid /// creating a shuffle type that targets may not be able to lower effectively. static Instruction *shrinkSplatShuffle(TruncInst &Trunc, InstCombiner::BuilderTy &Builder) { auto *Shuf = dyn_cast(Trunc.getOperand(0)); if (Shuf && Shuf->hasOneUse() && match(Shuf->getOperand(1), m_Undef()) && is_splat(Shuf->getShuffleMask()) && Shuf->getType() == Shuf->getOperand(0)->getType()) { // trunc (shuf X, Undef, SplatMask) --> shuf (trunc X), Undef, SplatMask Constant *NarrowUndef = UndefValue::get(Trunc.getType()); Value *NarrowOp = Builder.CreateTrunc(Shuf->getOperand(0), Trunc.getType()); return new ShuffleVectorInst(NarrowOp, NarrowUndef, Shuf->getShuffleMask()); } return nullptr; } /// Try to narrow the width of an insert element. This could be generalized for /// any vector constant, but we limit the transform to insertion into undef to /// avoid potential backend problems from unsupported insertion widths. This /// could also be extended to handle the case of inserting a scalar constant /// into a vector variable. static Instruction *shrinkInsertElt(CastInst &Trunc, InstCombiner::BuilderTy &Builder) { Instruction::CastOps Opcode = Trunc.getOpcode(); assert((Opcode == Instruction::Trunc || Opcode == Instruction::FPTrunc) && "Unexpected instruction for shrinking"); auto *InsElt = dyn_cast(Trunc.getOperand(0)); if (!InsElt || !InsElt->hasOneUse()) return nullptr; Type *DestTy = Trunc.getType(); Type *DestScalarTy = DestTy->getScalarType(); Value *VecOp = InsElt->getOperand(0); Value *ScalarOp = InsElt->getOperand(1); Value *Index = InsElt->getOperand(2); if (match(VecOp, m_Undef())) { // trunc (inselt undef, X, Index) --> inselt undef, (trunc X), Index // fptrunc (inselt undef, X, Index) --> inselt undef, (fptrunc X), Index UndefValue *NarrowUndef = UndefValue::get(DestTy); Value *NarrowOp = Builder.CreateCast(Opcode, ScalarOp, DestScalarTy); return InsertElementInst::Create(NarrowUndef, NarrowOp, Index); } return nullptr; } Instruction *InstCombinerImpl::visitTrunc(TruncInst &Trunc) { if (Instruction *Result = commonCastTransforms(Trunc)) return Result; Value *Src = Trunc.getOperand(0); Type *DestTy = Trunc.getType(), *SrcTy = Src->getType(); unsigned DestWidth = DestTy->getScalarSizeInBits(); unsigned SrcWidth = SrcTy->getScalarSizeInBits(); // Attempt to truncate the entire input expression tree to the destination // type. Only do this if the dest type is a simple type, don't convert the // expression tree to something weird like i93 unless the source is also // strange. if ((DestTy->isVectorTy() || shouldChangeType(SrcTy, DestTy)) && canEvaluateTruncated(Src, DestTy, *this, &Trunc)) { // If this cast is a truncate, evaluting in a different type always // eliminates the cast, so it is always a win. LLVM_DEBUG( dbgs() << "ICE: EvaluateInDifferentType converting expression type" " to avoid cast: " << Trunc << '\n'); Value *Res = EvaluateInDifferentType(Src, DestTy, false); assert(Res->getType() == DestTy); return replaceInstUsesWith(Trunc, Res); } // For integer types, check if we can shorten the entire input expression to // DestWidth * 2, which won't allow removing the truncate, but reducing the // width may enable further optimizations, e.g. allowing for larger // vectorization factors. if (auto *DestITy = dyn_cast(DestTy)) { if (DestWidth * 2 < SrcWidth) { auto *NewDestTy = DestITy->getExtendedType(); if (shouldChangeType(SrcTy, NewDestTy) && canEvaluateTruncated(Src, NewDestTy, *this, &Trunc)) { LLVM_DEBUG( dbgs() << "ICE: EvaluateInDifferentType converting expression type" " to reduce the width of operand of" << Trunc << '\n'); Value *Res = EvaluateInDifferentType(Src, NewDestTy, false); return new TruncInst(Res, DestTy); } } } // Test if the trunc is the user of a select which is part of a // minimum or maximum operation. If so, don't do any more simplification. // Even simplifying demanded bits can break the canonical form of a // min/max. Value *LHS, *RHS; if (SelectInst *Sel = dyn_cast(Src)) if (matchSelectPattern(Sel, LHS, RHS).Flavor != SPF_UNKNOWN) return nullptr; // See if we can simplify any instructions used by the input whose sole // purpose is to compute bits we don't care about. if (SimplifyDemandedInstructionBits(Trunc)) return &Trunc; if (DestWidth == 1) { Value *Zero = Constant::getNullValue(SrcTy); 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; Constant *C; if (match(Src, m_OneUse(m_LShr(m_Value(X), m_Constant(C))))) { // trunc (lshr X, C) to i1 --> icmp ne (and X, C'), 0 Constant *One = ConstantInt::get(SrcTy, APInt(SrcWidth, 1)); Constant *MaskC = ConstantExpr::getShl(One, C); Value *And = Builder.CreateAnd(X, MaskC); return new ICmpInst(ICmpInst::ICMP_NE, And, Zero); } if (match(Src, m_OneUse(m_c_Or(m_LShr(m_Value(X), m_Constant(C)), m_Deferred(X))))) { // trunc (or (lshr X, C), X) to i1 --> icmp ne (and X, C'), 0 Constant *One = ConstantInt::get(SrcTy, APInt(SrcWidth, 1)); Constant *MaskC = ConstantExpr::getShl(One, C); MaskC = ConstantExpr::getOr(MaskC, One); Value *And = Builder.CreateAnd(X, MaskC); return new ICmpInst(ICmpInst::ICMP_NE, And, Zero); } } Value *A, *B; Constant *C; if (match(Src, m_LShr(m_SExt(m_Value(A)), m_Constant(C)))) { unsigned AWidth = A->getType()->getScalarSizeInBits(); unsigned MaxShiftAmt = SrcWidth - std::max(DestWidth, AWidth); auto *OldSh = cast(Src); bool IsExact = OldSh->isExact(); // If the shift is small enough, all zero bits created by the shift are // removed by the trunc. if (match(C, m_SpecificInt_ICMP(ICmpInst::ICMP_ULE, APInt(SrcWidth, MaxShiftAmt)))) { // trunc (lshr (sext A), C) --> ashr A, C if (A->getType() == DestTy) { Constant *MaxAmt = ConstantInt::get(SrcTy, DestWidth - 1, false); Constant *ShAmt = ConstantExpr::getUMin(C, MaxAmt); ShAmt = ConstantExpr::getTrunc(ShAmt, A->getType()); ShAmt = Constant::mergeUndefsWith(ShAmt, C); return IsExact ? BinaryOperator::CreateExactAShr(A, ShAmt) : BinaryOperator::CreateAShr(A, ShAmt); } // The types are mismatched, so create a cast after shifting: // trunc (lshr (sext A), C) --> sext/trunc (ashr A, C) if (Src->hasOneUse()) { Constant *MaxAmt = ConstantInt::get(SrcTy, AWidth - 1, false); Constant *ShAmt = ConstantExpr::getUMin(C, MaxAmt); ShAmt = ConstantExpr::getTrunc(ShAmt, A->getType()); Value *Shift = Builder.CreateAShr(A, ShAmt, "", IsExact); return CastInst::CreateIntegerCast(Shift, DestTy, true); } } // TODO: Mask high bits with 'and'. } // trunc (*shr (trunc A), C) --> trunc(*shr A, C) if (match(Src, m_OneUse(m_Shr(m_Trunc(m_Value(A)), m_Constant(C))))) { unsigned MaxShiftAmt = SrcWidth - DestWidth; // If the shift is small enough, all zero/sign bits created by the shift are // removed by the trunc. if (match(C, m_SpecificInt_ICMP(ICmpInst::ICMP_ULE, APInt(SrcWidth, MaxShiftAmt)))) { auto *OldShift = cast(Src); bool IsExact = OldShift->isExact(); auto *ShAmt = ConstantExpr::getIntegerCast(C, A->getType(), true); ShAmt = Constant::mergeUndefsWith(ShAmt, C); Value *Shift = OldShift->getOpcode() == Instruction::AShr ? Builder.CreateAShr(A, ShAmt, OldShift->getName(), IsExact) : Builder.CreateLShr(A, ShAmt, OldShift->getName(), IsExact); return CastInst::CreateTruncOrBitCast(Shift, DestTy); } } if (Instruction *I = narrowBinOp(Trunc)) return I; if (Instruction *I = shrinkSplatShuffle(Trunc, Builder)) return I; if (Instruction *I = shrinkInsertElt(Trunc, Builder)) return I; if (Src->hasOneUse() && (isa(SrcTy) || shouldChangeType(SrcTy, DestTy))) { // Transform "trunc (shl X, cst)" -> "shl (trunc X), cst" so long as the // dest type is native and cst < dest size. if (match(Src, m_Shl(m_Value(A), m_Constant(C))) && !match(A, m_Shr(m_Value(), m_Constant()))) { // Skip shifts of shift by constants. It undoes a combine in // FoldShiftByConstant and is the extend in reg pattern. APInt Threshold = APInt(C->getType()->getScalarSizeInBits(), DestWidth); if (match(C, m_SpecificInt_ICMP(ICmpInst::ICMP_ULT, Threshold))) { Value *NewTrunc = Builder.CreateTrunc(A, DestTy, A->getName() + ".tr"); return BinaryOperator::Create(Instruction::Shl, NewTrunc, ConstantExpr::getTrunc(C, DestTy)); } } } if (Instruction *I = foldVecTruncToExtElt(Trunc, *this)) return I; // Whenever an element is extracted from a vector, and then truncated, // canonicalize by converting it to a bitcast followed by an // extractelement. // // Example (little endian): // trunc (extractelement <4 x i64> %X, 0) to i32 // ---> // extractelement <8 x i32> (bitcast <4 x i64> %X to <8 x i32>), i32 0 Value *VecOp; ConstantInt *Cst; if (match(Src, m_OneUse(m_ExtractElt(m_Value(VecOp), m_ConstantInt(Cst))))) { auto *VecOpTy = cast(VecOp->getType()); auto VecElts = VecOpTy->getElementCount(); // A badly fit destination size would result in an invalid cast. if (SrcWidth % DestWidth == 0) { uint64_t TruncRatio = SrcWidth / DestWidth; uint64_t BitCastNumElts = VecElts.getKnownMinValue() * TruncRatio; uint64_t VecOpIdx = Cst->getZExtValue(); uint64_t NewIdx = DL.isBigEndian() ? (VecOpIdx + 1) * TruncRatio - 1 : VecOpIdx * TruncRatio; assert(BitCastNumElts <= std::numeric_limits::max() && "overflow 32-bits"); auto *BitCastTo = VectorType::get(DestTy, BitCastNumElts, VecElts.isScalable()); Value *BitCast = Builder.CreateBitCast(VecOp, BitCastTo); return ExtractElementInst::Create(BitCast, Builder.getInt32(NewIdx)); } } // trunc (ctlz_i32(zext(A), B) --> add(ctlz_i16(A, B), C) if (match(Src, m_OneUse(m_Intrinsic(m_ZExt(m_Value(A)), m_Value(B))))) { unsigned AWidth = A->getType()->getScalarSizeInBits(); if (AWidth == DestWidth && AWidth > Log2_32(SrcWidth)) { Value *WidthDiff = ConstantInt::get(A->getType(), SrcWidth - AWidth); Value *NarrowCtlz = Builder.CreateIntrinsic(Intrinsic::ctlz, {Trunc.getType()}, {A, B}); return BinaryOperator::CreateAdd(NarrowCtlz, WidthDiff); } } return nullptr; } /// Transform (zext icmp) to bitwise / integer operations in order to /// eliminate it. If DoTransform is false, just test whether the given /// (zext icmp) can be transformed. Instruction *InstCombinerImpl::transformZExtICmp(ICmpInst *Cmp, ZExtInst &Zext, bool DoTransform) { // If we are just checking for a icmp eq of a single bit and zext'ing it // to an integer, then shift the bit to the appropriate place and then // cast to integer to avoid the comparison. const APInt *Op1CV; if (match(Cmp->getOperand(1), m_APInt(Op1CV))) { // zext (x x>>u31 true if signbit set. // zext (x >s -1) to i32 --> (x>>u31)^1 true if signbit clear. if ((Cmp->getPredicate() == ICmpInst::ICMP_SLT && Op1CV->isNullValue()) || (Cmp->getPredicate() == ICmpInst::ICMP_SGT && Op1CV->isAllOnesValue())) { if (!DoTransform) return Cmp; Value *In = Cmp->getOperand(0); Value *Sh = ConstantInt::get(In->getType(), In->getType()->getScalarSizeInBits() - 1); In = Builder.CreateLShr(In, Sh, In->getName() + ".lobit"); if (In->getType() != Zext.getType()) In = Builder.CreateIntCast(In, Zext.getType(), false /*ZExt*/); if (Cmp->getPredicate() == ICmpInst::ICMP_SGT) { Constant *One = ConstantInt::get(In->getType(), 1); In = Builder.CreateXor(In, One, In->getName() + ".not"); } return replaceInstUsesWith(Zext, In); } // zext (X == 0) to i32 --> X^1 iff X has only the low bit set. // zext (X == 0) to i32 --> (X>>1)^1 iff X has only the 2nd bit set. // zext (X == 1) to i32 --> X iff X has only the low bit set. // zext (X == 2) to i32 --> X>>1 iff X has only the 2nd bit set. // zext (X != 0) to i32 --> X iff X has only the low bit set. // zext (X != 0) to i32 --> X>>1 iff X has only the 2nd bit set. // zext (X != 1) to i32 --> X^1 iff X has only the low bit set. // zext (X != 2) to i32 --> (X>>1)^1 iff X has only the 2nd bit set. if ((Op1CV->isNullValue() || Op1CV->isPowerOf2()) && // This only works for EQ and NE Cmp->isEquality()) { // If Op1C some other power of two, convert: KnownBits Known = computeKnownBits(Cmp->getOperand(0), 0, &Zext); APInt KnownZeroMask(~Known.Zero); if (KnownZeroMask.isPowerOf2()) { // Exactly 1 possible 1? if (!DoTransform) return Cmp; bool isNE = Cmp->getPredicate() == ICmpInst::ICMP_NE; if (!Op1CV->isNullValue() && (*Op1CV != KnownZeroMask)) { // (X&4) == 2 --> false // (X&4) != 2 --> true Constant *Res = ConstantInt::get(Zext.getType(), isNE); return replaceInstUsesWith(Zext, Res); } uint32_t ShAmt = KnownZeroMask.logBase2(); Value *In = Cmp->getOperand(0); if (ShAmt) { // Perform a logical shr by shiftamt. // Insert the shift to put the result in the low bit. In = Builder.CreateLShr(In, ConstantInt::get(In->getType(), ShAmt), In->getName() + ".lobit"); } if (!Op1CV->isNullValue() == isNE) { // Toggle the low bit. Constant *One = ConstantInt::get(In->getType(), 1); In = Builder.CreateXor(In, One); } if (Zext.getType() == In->getType()) return replaceInstUsesWith(Zext, In); Value *IntCast = Builder.CreateIntCast(In, Zext.getType(), false); return replaceInstUsesWith(Zext, IntCast); } } } if (Cmp->isEquality() && Zext.getType() == Cmp->getOperand(0)->getType()) { // Test if a bit is clear/set using a shifted-one mask: // zext (icmp eq (and X, (1 << ShAmt)), 0) --> and (lshr (not X), ShAmt), 1 // zext (icmp ne (and X, (1 << ShAmt)), 0) --> and (lshr X, ShAmt), 1 Value *X, *ShAmt; if (Cmp->hasOneUse() && match(Cmp->getOperand(1), m_ZeroInt()) && match(Cmp->getOperand(0), m_OneUse(m_c_And(m_Shl(m_One(), m_Value(ShAmt)), m_Value(X))))) { if (!DoTransform) return Cmp; if (Cmp->getPredicate() == ICmpInst::ICMP_EQ) X = Builder.CreateNot(X); Value *Lshr = Builder.CreateLShr(X, ShAmt); Value *And1 = Builder.CreateAnd(Lshr, ConstantInt::get(X->getType(), 1)); return replaceInstUsesWith(Zext, And1); } // icmp ne A, B is equal to xor A, B when A and B only really have one bit. // It is also profitable to transform icmp eq into not(xor(A, B)) because // that may lead to additional simplifications. if (IntegerType *ITy = dyn_cast(Zext.getType())) { Value *LHS = Cmp->getOperand(0); Value *RHS = Cmp->getOperand(1); KnownBits KnownLHS = computeKnownBits(LHS, 0, &Zext); KnownBits KnownRHS = computeKnownBits(RHS, 0, &Zext); if (KnownLHS.Zero == KnownRHS.Zero && KnownLHS.One == KnownRHS.One) { APInt KnownBits = KnownLHS.Zero | KnownLHS.One; APInt UnknownBit = ~KnownBits; if (UnknownBit.countPopulation() == 1) { if (!DoTransform) return Cmp; Value *Result = Builder.CreateXor(LHS, RHS); // Mask off any bits that are set and won't be shifted away. if (KnownLHS.One.uge(UnknownBit)) Result = Builder.CreateAnd(Result, ConstantInt::get(ITy, UnknownBit)); // Shift the bit we're testing down to the lsb. Result = Builder.CreateLShr( Result, ConstantInt::get(ITy, UnknownBit.countTrailingZeros())); if (Cmp->getPredicate() == ICmpInst::ICMP_EQ) Result = Builder.CreateXor(Result, ConstantInt::get(ITy, 1)); Result->takeName(Cmp); return replaceInstUsesWith(Zext, Result); } } } } return nullptr; } /// Determine if the specified value can be computed in the specified wider type /// and produce the same low bits. If not, return false. /// /// If this function returns true, it can also return a non-zero number of bits /// (in BitsToClear) which indicates that the value it computes is correct for /// the zero extend, but that the additional BitsToClear bits need to be zero'd /// out. For example, to promote something like: /// /// %B = trunc i64 %A to i32 /// %C = lshr i32 %B, 8 /// %E = zext i32 %C to i64 /// /// CanEvaluateZExtd for the 'lshr' will return true, and BitsToClear will be /// set to 8 to indicate that the promoted value needs to have bits 24-31 /// cleared in addition to bits 32-63. Since an 'and' will be generated to /// clear the top bits anyway, doing this has no extra cost. /// /// This function works on both vectors and scalars. static bool canEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear, InstCombinerImpl &IC, Instruction *CxtI) { BitsToClear = 0; if (canAlwaysEvaluateInType(V, Ty)) return true; if (canNotEvaluateInType(V, Ty)) return false; auto *I = cast(V); unsigned Tmp; switch (I->getOpcode()) { case Instruction::ZExt: // zext(zext(x)) -> zext(x). case Instruction::SExt: // zext(sext(x)) -> sext(x). case Instruction::Trunc: // zext(trunc(x)) -> trunc(x) or zext(x) return true; case Instruction::And: case Instruction::Or: case Instruction::Xor: case Instruction::Add: case Instruction::Sub: case Instruction::Mul: if (!canEvaluateZExtd(I->getOperand(0), Ty, BitsToClear, IC, CxtI) || !canEvaluateZExtd(I->getOperand(1), Ty, Tmp, IC, CxtI)) return false; // These can all be promoted if neither operand has 'bits to clear'. if (BitsToClear == 0 && Tmp == 0) return true; // If the operation is an AND/OR/XOR and the bits to clear are zero in the // other side, BitsToClear is ok. if (Tmp == 0 && I->isBitwiseLogicOp()) { // We use MaskedValueIsZero here for generality, but the case we care // about the most is constant RHS. unsigned VSize = V->getType()->getScalarSizeInBits(); if (IC.MaskedValueIsZero(I->getOperand(1), APInt::getHighBitsSet(VSize, BitsToClear), 0, CxtI)) { // If this is an And instruction and all of the BitsToClear are // known to be zero we can reset BitsToClear. if (I->getOpcode() == Instruction::And) BitsToClear = 0; return true; } } // Otherwise, we don't know how to analyze this BitsToClear case yet. return false; case Instruction::Shl: { // We can promote shl(x, cst) if we can promote x. Since shl overwrites the // upper bits we can reduce BitsToClear by the shift amount. const APInt *Amt; if (match(I->getOperand(1), m_APInt(Amt))) { if (!canEvaluateZExtd(I->getOperand(0), Ty, BitsToClear, IC, CxtI)) return false; uint64_t ShiftAmt = Amt->getZExtValue(); BitsToClear = ShiftAmt < BitsToClear ? BitsToClear - ShiftAmt : 0; return true; } return false; } case Instruction::LShr: { // We can promote lshr(x, cst) if we can promote x. This requires the // ultimate 'and' to clear out the high zero bits we're clearing out though. const APInt *Amt; if (match(I->getOperand(1), m_APInt(Amt))) { if (!canEvaluateZExtd(I->getOperand(0), Ty, BitsToClear, IC, CxtI)) return false; BitsToClear += Amt->getZExtValue(); if (BitsToClear > V->getType()->getScalarSizeInBits()) BitsToClear = V->getType()->getScalarSizeInBits(); return true; } // Cannot promote variable LSHR. return false; } case Instruction::Select: if (!canEvaluateZExtd(I->getOperand(1), Ty, Tmp, IC, CxtI) || !canEvaluateZExtd(I->getOperand(2), Ty, BitsToClear, IC, CxtI) || // TODO: If important, we could handle the case when the BitsToClear are // known zero in the disagreeing side. Tmp != BitsToClear) return false; return true; case Instruction::PHI: { // We can change a phi if we can change all operands. Note that we never // get into trouble with cyclic PHIs here because we only consider // instructions with a single use. PHINode *PN = cast(I); if (!canEvaluateZExtd(PN->getIncomingValue(0), Ty, BitsToClear, IC, CxtI)) return false; for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) if (!canEvaluateZExtd(PN->getIncomingValue(i), Ty, Tmp, IC, CxtI) || // TODO: If important, we could handle the case when the BitsToClear // are known zero in the disagreeing input. Tmp != BitsToClear) return false; return true; } default: // TODO: Can handle more cases here. return false; } } Instruction *InstCombinerImpl::visitZExt(ZExtInst &CI) { // If this zero extend is only used by a truncate, let the truncate be // eliminated before we try to optimize this zext. if (CI.hasOneUse() && isa(CI.user_back())) return nullptr; // If one of the common conversion will work, do it. if (Instruction *Result = commonCastTransforms(CI)) return Result; Value *Src = CI.getOperand(0); Type *SrcTy = Src->getType(), *DestTy = CI.getType(); // Try to extend the entire expression tree to the wide destination type. unsigned BitsToClear; if (shouldChangeType(SrcTy, DestTy) && canEvaluateZExtd(Src, DestTy, BitsToClear, *this, &CI)) { assert(BitsToClear <= SrcTy->getScalarSizeInBits() && "Can't clear more bits than in SrcTy"); // Okay, we can transform this! Insert the new expression now. LLVM_DEBUG( dbgs() << "ICE: EvaluateInDifferentType converting expression type" " to avoid zero extend: " << CI << '\n'); Value *Res = EvaluateInDifferentType(Src, DestTy, false); assert(Res->getType() == DestTy); // Preserve debug values referring to Src if the zext is its last use. if (auto *SrcOp = dyn_cast(Src)) if (SrcOp->hasOneUse()) replaceAllDbgUsesWith(*SrcOp, *Res, CI, DT); uint32_t SrcBitsKept = SrcTy->getScalarSizeInBits()-BitsToClear; uint32_t DestBitSize = DestTy->getScalarSizeInBits(); // If the high bits are already filled with zeros, just replace this // cast with the result. if (MaskedValueIsZero(Res, APInt::getHighBitsSet(DestBitSize, DestBitSize-SrcBitsKept), 0, &CI)) return replaceInstUsesWith(CI, Res); // We need to emit an AND to clear the high bits. Constant *C = ConstantInt::get(Res->getType(), APInt::getLowBitsSet(DestBitSize, SrcBitsKept)); return BinaryOperator::CreateAnd(Res, C); } // If this is a TRUNC followed by a ZEXT then we are dealing with integral // types and if the sizes are just right we can convert this into a logical // 'and' which will be much cheaper than the pair of casts. if (TruncInst *CSrc = dyn_cast(Src)) { // A->B->C cast // TODO: Subsume this into EvaluateInDifferentType. // Get the sizes of the types involved. We know that the intermediate type // will be smaller than A or C, but don't know the relation between A and C. Value *A = CSrc->getOperand(0); unsigned SrcSize = A->getType()->getScalarSizeInBits(); unsigned MidSize = CSrc->getType()->getScalarSizeInBits(); unsigned DstSize = CI.getType()->getScalarSizeInBits(); // If we're actually extending zero bits, then if // SrcSize < DstSize: zext(a & mask) // SrcSize == DstSize: a & mask // SrcSize > DstSize: trunc(a) & mask if (SrcSize < DstSize) { APInt AndValue(APInt::getLowBitsSet(SrcSize, MidSize)); Constant *AndConst = ConstantInt::get(A->getType(), AndValue); Value *And = Builder.CreateAnd(A, AndConst, CSrc->getName() + ".mask"); return new ZExtInst(And, CI.getType()); } if (SrcSize == DstSize) { APInt AndValue(APInt::getLowBitsSet(SrcSize, MidSize)); return BinaryOperator::CreateAnd(A, ConstantInt::get(A->getType(), AndValue)); } if (SrcSize > DstSize) { Value *Trunc = Builder.CreateTrunc(A, CI.getType()); APInt AndValue(APInt::getLowBitsSet(DstSize, MidSize)); return BinaryOperator::CreateAnd(Trunc, ConstantInt::get(Trunc->getType(), AndValue)); } } if (ICmpInst *Cmp = dyn_cast(Src)) return transformZExtICmp(Cmp, CI); BinaryOperator *SrcI = dyn_cast(Src); if (SrcI && SrcI->getOpcode() == Instruction::Or) { // zext (or icmp, icmp) -> or (zext icmp), (zext icmp) if at least one // of the (zext icmp) can be eliminated. If so, immediately perform the // according elimination. ICmpInst *LHS = dyn_cast(SrcI->getOperand(0)); ICmpInst *RHS = dyn_cast(SrcI->getOperand(1)); if (LHS && RHS && LHS->hasOneUse() && RHS->hasOneUse() && LHS->getOperand(0)->getType() == RHS->getOperand(0)->getType() && (transformZExtICmp(LHS, CI, false) || transformZExtICmp(RHS, CI, false))) { // zext (or icmp, icmp) -> or (zext icmp), (zext icmp) Value *LCast = Builder.CreateZExt(LHS, CI.getType(), LHS->getName()); Value *RCast = Builder.CreateZExt(RHS, CI.getType(), RHS->getName()); Value *Or = Builder.CreateOr(LCast, RCast, CI.getName()); if (auto *OrInst = dyn_cast(Or)) Builder.SetInsertPoint(OrInst); // Perform the elimination. if (auto *LZExt = dyn_cast(LCast)) transformZExtICmp(LHS, *LZExt); if (auto *RZExt = dyn_cast(RCast)) transformZExtICmp(RHS, *RZExt); return replaceInstUsesWith(CI, Or); } } // zext(trunc(X) & C) -> (X & zext(C)). Constant *C; Value *X; if (SrcI && match(SrcI, m_OneUse(m_And(m_Trunc(m_Value(X)), m_Constant(C)))) && X->getType() == CI.getType()) return BinaryOperator::CreateAnd(X, ConstantExpr::getZExt(C, CI.getType())); // zext((trunc(X) & C) ^ C) -> ((X & zext(C)) ^ zext(C)). Value *And; if (SrcI && match(SrcI, m_OneUse(m_Xor(m_Value(And), m_Constant(C)))) && match(And, m_OneUse(m_And(m_Trunc(m_Value(X)), m_Specific(C)))) && X->getType() == CI.getType()) { Constant *ZC = ConstantExpr::getZExt(C, CI.getType()); return BinaryOperator::CreateXor(Builder.CreateAnd(X, ZC), ZC); } return nullptr; } /// Transform (sext icmp) to bitwise / integer operations to eliminate the icmp. Instruction *InstCombinerImpl::transformSExtICmp(ICmpInst *ICI, Instruction &CI) { Value *Op0 = ICI->getOperand(0), *Op1 = ICI->getOperand(1); ICmpInst::Predicate Pred = ICI->getPredicate(); // Don't bother if Op1 isn't of vector or integer type. if (!Op1->getType()->isIntOrIntVectorTy()) return nullptr; if ((Pred == ICmpInst::ICMP_SLT && match(Op1, m_ZeroInt())) || (Pred == ICmpInst::ICMP_SGT && match(Op1, m_AllOnes()))) { // (x ashr x, 31 -> all ones if negative // (x >s -1) ? -1 : 0 -> not (ashr x, 31) -> all ones if positive Value *Sh = ConstantInt::get(Op0->getType(), Op0->getType()->getScalarSizeInBits() - 1); Value *In = Builder.CreateAShr(Op0, Sh, Op0->getName() + ".lobit"); if (In->getType() != CI.getType()) In = Builder.CreateIntCast(In, CI.getType(), true /*SExt*/); if (Pred == ICmpInst::ICMP_SGT) In = Builder.CreateNot(In, In->getName() + ".not"); return replaceInstUsesWith(CI, In); } if (ConstantInt *Op1C = dyn_cast(Op1)) { // If we know that only one bit of the LHS of the icmp can be set and we // have an equality comparison with zero or a power of 2, we can transform // the icmp and sext into bitwise/integer operations. if (ICI->hasOneUse() && ICI->isEquality() && (Op1C->isZero() || Op1C->getValue().isPowerOf2())){ KnownBits Known = computeKnownBits(Op0, 0, &CI); APInt KnownZeroMask(~Known.Zero); if (KnownZeroMask.isPowerOf2()) { Value *In = ICI->getOperand(0); // If the icmp tests for a known zero bit we can constant fold it. if (!Op1C->isZero() && Op1C->getValue() != KnownZeroMask) { Value *V = Pred == ICmpInst::ICMP_NE ? ConstantInt::getAllOnesValue(CI.getType()) : ConstantInt::getNullValue(CI.getType()); return replaceInstUsesWith(CI, V); } if (!Op1C->isZero() == (Pred == ICmpInst::ICMP_NE)) { // sext ((x & 2^n) == 0) -> (x >> n) - 1 // sext ((x & 2^n) != 2^n) -> (x >> n) - 1 unsigned ShiftAmt = KnownZeroMask.countTrailingZeros(); // Perform a right shift to place the desired bit in the LSB. if (ShiftAmt) In = Builder.CreateLShr(In, ConstantInt::get(In->getType(), ShiftAmt)); // At this point "In" is either 1 or 0. Subtract 1 to turn // {1, 0} -> {0, -1}. In = Builder.CreateAdd(In, ConstantInt::getAllOnesValue(In->getType()), "sext"); } else { // sext ((x & 2^n) != 0) -> (x << bitwidth-n) a>> bitwidth-1 // sext ((x & 2^n) == 2^n) -> (x << bitwidth-n) a>> bitwidth-1 unsigned ShiftAmt = KnownZeroMask.countLeadingZeros(); // Perform a left shift to place the desired bit in the MSB. if (ShiftAmt) In = Builder.CreateShl(In, ConstantInt::get(In->getType(), ShiftAmt)); // Distribute the bit over the whole bit width. In = Builder.CreateAShr(In, ConstantInt::get(In->getType(), KnownZeroMask.getBitWidth() - 1), "sext"); } if (CI.getType() == In->getType()) return replaceInstUsesWith(CI, In); return CastInst::CreateIntegerCast(In, CI.getType(), true/*SExt*/); } } } return nullptr; } /// Return true if we can take the specified value and return it as type Ty /// without inserting any new casts and without changing the value of the common /// low bits. This is used by code that tries to promote integer operations to /// a wider types will allow us to eliminate the extension. /// /// This function works on both vectors and scalars. /// static bool canEvaluateSExtd(Value *V, Type *Ty) { assert(V->getType()->getScalarSizeInBits() < Ty->getScalarSizeInBits() && "Can't sign extend type to a smaller type"); if (canAlwaysEvaluateInType(V, Ty)) return true; if (canNotEvaluateInType(V, Ty)) return false; auto *I = cast(V); switch (I->getOpcode()) { case Instruction::SExt: // sext(sext(x)) -> sext(x) case Instruction::ZExt: // sext(zext(x)) -> zext(x) case Instruction::Trunc: // sext(trunc(x)) -> trunc(x) or sext(x) return true; case Instruction::And: case Instruction::Or: case Instruction::Xor: case Instruction::Add: case Instruction::Sub: case Instruction::Mul: // These operators can all arbitrarily be extended if their inputs can. return canEvaluateSExtd(I->getOperand(0), Ty) && canEvaluateSExtd(I->getOperand(1), Ty); //case Instruction::Shl: TODO //case Instruction::LShr: TODO case Instruction::Select: return canEvaluateSExtd(I->getOperand(1), Ty) && canEvaluateSExtd(I->getOperand(2), Ty); case Instruction::PHI: { // We can change a phi if we can change all operands. Note that we never // get into trouble with cyclic PHIs here because we only consider // instructions with a single use. PHINode *PN = cast(I); for (Value *IncValue : PN->incoming_values()) if (!canEvaluateSExtd(IncValue, Ty)) return false; return true; } default: // TODO: Can handle more cases here. break; } return false; } Instruction *InstCombinerImpl::visitSExt(SExtInst &CI) { // If this sign extend is only used by a truncate, let the truncate be // eliminated before we try to optimize this sext. if (CI.hasOneUse() && isa(CI.user_back())) return nullptr; if (Instruction *I = commonCastTransforms(CI)) return I; Value *Src = CI.getOperand(0); Type *SrcTy = Src->getType(), *DestTy = CI.getType(); unsigned SrcBitSize = SrcTy->getScalarSizeInBits(); unsigned DestBitSize = DestTy->getScalarSizeInBits(); // If we know that the value being extended is positive, we can use a zext // instead. KnownBits Known = computeKnownBits(Src, 0, &CI); if (Known.isNonNegative()) return CastInst::Create(Instruction::ZExt, Src, DestTy); // Try to extend the entire expression tree to the wide destination type. if (shouldChangeType(SrcTy, DestTy) && canEvaluateSExtd(Src, DestTy)) { // Okay, we can transform this! Insert the new expression now. LLVM_DEBUG( dbgs() << "ICE: EvaluateInDifferentType converting expression type" " to avoid sign extend: " << CI << '\n'); Value *Res = EvaluateInDifferentType(Src, DestTy, true); assert(Res->getType() == DestTy); // If the high bits are already filled with sign bit, just replace this // cast with the result. if (ComputeNumSignBits(Res, 0, &CI) > DestBitSize - SrcBitSize) return replaceInstUsesWith(CI, Res); // We need to emit a shl + ashr to do the sign extend. Value *ShAmt = ConstantInt::get(DestTy, DestBitSize-SrcBitSize); return BinaryOperator::CreateAShr(Builder.CreateShl(Res, ShAmt, "sext"), ShAmt); } Value *X; if (match(Src, m_Trunc(m_Value(X)))) { // If the input has more sign bits than bits truncated, then convert // directly to final type. unsigned XBitSize = X->getType()->getScalarSizeInBits(); if (ComputeNumSignBits(X, 0, &CI) > XBitSize - SrcBitSize) return CastInst::CreateIntegerCast(X, DestTy, /* isSigned */ true); // If input is a trunc from the destination type, then convert into shifts. if (Src->hasOneUse() && X->getType() == DestTy) { // sext (trunc X) --> ashr (shl X, C), C Constant *ShAmt = ConstantInt::get(DestTy, DestBitSize - SrcBitSize); return BinaryOperator::CreateAShr(Builder.CreateShl(X, ShAmt), ShAmt); } // If we are replacing shifted-in high zero bits with sign bits, convert // the logic shift to arithmetic shift and eliminate the cast to // intermediate type: // sext (trunc (lshr Y, C)) --> sext/trunc (ashr Y, C) Value *Y; if (Src->hasOneUse() && match(X, m_LShr(m_Value(Y), m_SpecificIntAllowUndef(XBitSize - SrcBitSize)))) { Value *Ashr = Builder.CreateAShr(Y, XBitSize - SrcBitSize); return CastInst::CreateIntegerCast(Ashr, DestTy, /* isSigned */ true); } } if (ICmpInst *ICI = dyn_cast(Src)) return transformSExtICmp(ICI, CI); // If the input is a shl/ashr pair of a same constant, then this is a sign // extension from a smaller value. If we could trust arbitrary bitwidth // integers, we could turn this into a truncate to the smaller bit and then // use a sext for the whole extension. Since we don't, look deeper and check // for a truncate. If the source and dest are the same type, eliminate the // trunc and extend and just do shifts. For example, turn: // %a = trunc i32 %i to i8 // %b = shl i8 %a, C // %c = ashr i8 %b, C // %d = sext i8 %c to i32 // into: // %a = shl i32 %i, 32-(8-C) // %d = ashr i32 %a, 32-(8-C) Value *A = nullptr; // TODO: Eventually this could be subsumed by EvaluateInDifferentType. Constant *BA = nullptr, *CA = nullptr; if (match(Src, m_AShr(m_Shl(m_Trunc(m_Value(A)), m_Constant(BA)), m_Constant(CA))) && BA->isElementWiseEqual(CA) && A->getType() == DestTy) { Constant *WideCurrShAmt = ConstantExpr::getSExt(CA, DestTy); Constant *NumLowbitsLeft = ConstantExpr::getSub( ConstantInt::get(DestTy, SrcTy->getScalarSizeInBits()), WideCurrShAmt); Constant *NewShAmt = ConstantExpr::getSub( ConstantInt::get(DestTy, DestTy->getScalarSizeInBits()), NumLowbitsLeft); NewShAmt = Constant::mergeUndefsWith(Constant::mergeUndefsWith(NewShAmt, BA), CA); A = Builder.CreateShl(A, NewShAmt, CI.getName()); return BinaryOperator::CreateAShr(A, NewShAmt); } return nullptr; } /// Return a Constant* for the specified floating-point constant if it fits /// in the specified FP type without changing its value. static bool fitsInFPType(ConstantFP *CFP, const fltSemantics &Sem) { bool losesInfo; APFloat F = CFP->getValueAPF(); (void)F.convert(Sem, APFloat::rmNearestTiesToEven, &losesInfo); return !losesInfo; } static Type *shrinkFPConstant(ConstantFP *CFP) { if (CFP->getType() == Type::getPPC_FP128Ty(CFP->getContext())) return nullptr; // No constant folding of this. // See if the value can be truncated to half and then reextended. if (fitsInFPType(CFP, APFloat::IEEEhalf())) return Type::getHalfTy(CFP->getContext()); // See if the value can be truncated to float and then reextended. if (fitsInFPType(CFP, APFloat::IEEEsingle())) return Type::getFloatTy(CFP->getContext()); if (CFP->getType()->isDoubleTy()) return nullptr; // Won't shrink. if (fitsInFPType(CFP, APFloat::IEEEdouble())) return Type::getDoubleTy(CFP->getContext()); // Don't try to shrink to various long double types. return nullptr; } // Determine if this is a vector of ConstantFPs and if so, return the minimal // type we can safely truncate all elements to. // TODO: Make these support undef elements. static Type *shrinkFPConstantVector(Value *V) { auto *CV = dyn_cast(V); auto *CVVTy = dyn_cast(V->getType()); if (!CV || !CVVTy) return nullptr; Type *MinType = nullptr; unsigned NumElts = CVVTy->getNumElements(); // For fixed-width vectors we find the minimal type by looking // through the constant values of the vector. for (unsigned i = 0; i != NumElts; ++i) { auto *CFP = dyn_cast_or_null(CV->getAggregateElement(i)); if (!CFP) return nullptr; Type *T = shrinkFPConstant(CFP); if (!T) return nullptr; // If we haven't found a type yet or this type has a larger mantissa than // our previous type, this is our new minimal type. if (!MinType || T->getFPMantissaWidth() > MinType->getFPMantissaWidth()) MinType = T; } // Make a vector type from the minimal type. return FixedVectorType::get(MinType, NumElts); } /// Find the minimum FP type we can safely truncate to. static Type *getMinimumFPType(Value *V) { if (auto *FPExt = dyn_cast(V)) return FPExt->getOperand(0)->getType(); // If this value is a constant, return the constant in the smallest FP type // that can accurately represent it. This allows us to turn // (float)((double)X+2.0) into x+2.0f. if (auto *CFP = dyn_cast(V)) if (Type *T = shrinkFPConstant(CFP)) return T; // We can only correctly find a minimum type for a scalable vector when it is // a splat. For splats of constant values the fpext is wrapped up as a // ConstantExpr. if (auto *FPCExt = dyn_cast(V)) if (FPCExt->getOpcode() == Instruction::FPExt) return FPCExt->getOperand(0)->getType(); // Try to shrink a vector of FP constants. This returns nullptr on scalable // vectors if (Type *T = shrinkFPConstantVector(V)) return T; return V->getType(); } /// Return true if the cast from integer to FP can be proven to be exact for all /// possible inputs (the conversion does not lose any precision). static bool isKnownExactCastIntToFP(CastInst &I) { CastInst::CastOps Opcode = I.getOpcode(); assert((Opcode == CastInst::SIToFP || Opcode == CastInst::UIToFP) && "Unexpected cast"); Value *Src = I.getOperand(0); Type *SrcTy = Src->getType(); Type *FPTy = I.getType(); bool IsSigned = Opcode == Instruction::SIToFP; int SrcSize = (int)SrcTy->getScalarSizeInBits() - IsSigned; // Easy case - if the source integer type has less bits than the FP mantissa, // then the cast must be exact. int DestNumSigBits = FPTy->getFPMantissaWidth(); if (SrcSize <= DestNumSigBits) return true; // Cast from FP to integer and back to FP is independent of the intermediate // integer width because of poison on overflow. Value *F; if (match(Src, m_FPToSI(m_Value(F))) || match(Src, m_FPToUI(m_Value(F)))) { // If this is uitofp (fptosi F), the source needs an extra bit to avoid // potential rounding of negative FP input values. int SrcNumSigBits = F->getType()->getFPMantissaWidth(); if (!IsSigned && match(Src, m_FPToSI(m_Value()))) SrcNumSigBits++; // [su]itofp (fpto[su]i F) --> exact if the source type has less or equal // significant bits than the destination (and make sure neither type is // weird -- ppc_fp128). if (SrcNumSigBits > 0 && DestNumSigBits > 0 && SrcNumSigBits <= DestNumSigBits) return true; } // TODO: // Try harder to find if the source integer type has less significant bits. // For example, compute number of sign bits or compute low bit mask. return false; } Instruction *InstCombinerImpl::visitFPTrunc(FPTruncInst &FPT) { if (Instruction *I = commonCastTransforms(FPT)) return I; // If we have fptrunc(OpI (fpextend x), (fpextend y)), we would like to // simplify this expression to avoid one or more of the trunc/extend // operations if we can do so without changing the numerical results. // // The exact manner in which the widths of the operands interact to limit // what we can and cannot do safely varies from operation to operation, and // is explained below in the various case statements. Type *Ty = FPT.getType(); auto *BO = dyn_cast(FPT.getOperand(0)); if (BO && BO->hasOneUse()) { Type *LHSMinType = getMinimumFPType(BO->getOperand(0)); Type *RHSMinType = getMinimumFPType(BO->getOperand(1)); unsigned OpWidth = BO->getType()->getFPMantissaWidth(); unsigned LHSWidth = LHSMinType->getFPMantissaWidth(); unsigned RHSWidth = RHSMinType->getFPMantissaWidth(); unsigned SrcWidth = std::max(LHSWidth, RHSWidth); unsigned DstWidth = Ty->getFPMantissaWidth(); switch (BO->getOpcode()) { default: break; case Instruction::FAdd: case Instruction::FSub: // For addition and subtraction, the infinitely precise result can // essentially be arbitrarily wide; proving that double rounding // will not occur because the result of OpI is exact (as we will for // FMul, for example) is hopeless. However, we *can* nonetheless // frequently know that double rounding cannot occur (or that it is // innocuous) by taking advantage of the specific structure of // infinitely-precise results that admit double rounding. // // Specifically, if OpWidth >= 2*DstWdith+1 and DstWidth is sufficient // to represent both sources, we can guarantee that the double // rounding is innocuous (See p50 of Figueroa's 2000 PhD thesis, // "A Rigorous Framework for Fully Supporting the IEEE Standard ..." // for proof of this fact). // // Note: Figueroa does not consider the case where DstFormat != // SrcFormat. It's possible (likely even!) that this analysis // could be tightened for those cases, but they are rare (the main // case of interest here is (float)((double)float + float)). if (OpWidth >= 2*DstWidth+1 && DstWidth >= SrcWidth) { Value *LHS = Builder.CreateFPTrunc(BO->getOperand(0), Ty); Value *RHS = Builder.CreateFPTrunc(BO->getOperand(1), Ty); Instruction *RI = BinaryOperator::Create(BO->getOpcode(), LHS, RHS); RI->copyFastMathFlags(BO); return RI; } break; case Instruction::FMul: // For multiplication, the infinitely precise result has at most // LHSWidth + RHSWidth significant bits; if OpWidth is sufficient // that such a value can be exactly represented, then no double // rounding can possibly occur; we can safely perform the operation // in the destination format if it can represent both sources. if (OpWidth >= LHSWidth + RHSWidth && DstWidth >= SrcWidth) { Value *LHS = Builder.CreateFPTrunc(BO->getOperand(0), Ty); Value *RHS = Builder.CreateFPTrunc(BO->getOperand(1), Ty); return BinaryOperator::CreateFMulFMF(LHS, RHS, BO); } break; case Instruction::FDiv: // For division, we use again use the bound from Figueroa's // dissertation. I am entirely certain that this bound can be // tightened in the unbalanced operand case by an analysis based on // the diophantine rational approximation bound, but the well-known // condition used here is a good conservative first pass. // TODO: Tighten bound via rigorous analysis of the unbalanced case. if (OpWidth >= 2*DstWidth && DstWidth >= SrcWidth) { Value *LHS = Builder.CreateFPTrunc(BO->getOperand(0), Ty); Value *RHS = Builder.CreateFPTrunc(BO->getOperand(1), Ty); return BinaryOperator::CreateFDivFMF(LHS, RHS, BO); } break; case Instruction::FRem: { // Remainder is straightforward. Remainder is always exact, so the // type of OpI doesn't enter into things at all. We simply evaluate // in whichever source type is larger, then convert to the // destination type. if (SrcWidth == OpWidth) break; Value *LHS, *RHS; if (LHSWidth == SrcWidth) { LHS = Builder.CreateFPTrunc(BO->getOperand(0), LHSMinType); RHS = Builder.CreateFPTrunc(BO->getOperand(1), LHSMinType); } else { LHS = Builder.CreateFPTrunc(BO->getOperand(0), RHSMinType); RHS = Builder.CreateFPTrunc(BO->getOperand(1), RHSMinType); } Value *ExactResult = Builder.CreateFRemFMF(LHS, RHS, BO); return CastInst::CreateFPCast(ExactResult, Ty); } } } // (fptrunc (fneg x)) -> (fneg (fptrunc x)) Value *X; Instruction *Op = dyn_cast(FPT.getOperand(0)); if (Op && Op->hasOneUse()) { // FIXME: The FMF should propagate from the fptrunc, not the source op. IRBuilder<>::FastMathFlagGuard FMFG(Builder); if (isa(Op)) Builder.setFastMathFlags(Op->getFastMathFlags()); if (match(Op, m_FNeg(m_Value(X)))) { Value *InnerTrunc = Builder.CreateFPTrunc(X, Ty); return UnaryOperator::CreateFNegFMF(InnerTrunc, Op); } // If we are truncating a select that has an extended operand, we can // narrow the other operand and do the select as a narrow op. Value *Cond, *X, *Y; if (match(Op, m_Select(m_Value(Cond), m_FPExt(m_Value(X)), m_Value(Y))) && X->getType() == Ty) { // fptrunc (select Cond, (fpext X), Y --> select Cond, X, (fptrunc Y) Value *NarrowY = Builder.CreateFPTrunc(Y, Ty); Value *Sel = Builder.CreateSelect(Cond, X, NarrowY, "narrow.sel", Op); return replaceInstUsesWith(FPT, Sel); } if (match(Op, m_Select(m_Value(Cond), m_Value(Y), m_FPExt(m_Value(X)))) && X->getType() == Ty) { // fptrunc (select Cond, Y, (fpext X) --> select Cond, (fptrunc Y), X Value *NarrowY = Builder.CreateFPTrunc(Y, Ty); Value *Sel = Builder.CreateSelect(Cond, NarrowY, X, "narrow.sel", Op); return replaceInstUsesWith(FPT, Sel); } } if (auto *II = dyn_cast(FPT.getOperand(0))) { switch (II->getIntrinsicID()) { default: break; case Intrinsic::ceil: case Intrinsic::fabs: case Intrinsic::floor: case Intrinsic::nearbyint: case Intrinsic::rint: case Intrinsic::round: case Intrinsic::roundeven: case Intrinsic::trunc: { Value *Src = II->getArgOperand(0); if (!Src->hasOneUse()) break; // Except for fabs, this transformation requires the input of the unary FP // operation to be itself an fpext from the type to which we're // truncating. if (II->getIntrinsicID() != Intrinsic::fabs) { FPExtInst *FPExtSrc = dyn_cast(Src); if (!FPExtSrc || FPExtSrc->getSrcTy() != Ty) break; } // Do unary FP operation on smaller type. // (fptrunc (fabs x)) -> (fabs (fptrunc x)) Value *InnerTrunc = Builder.CreateFPTrunc(Src, Ty); Function *Overload = Intrinsic::getDeclaration(FPT.getModule(), II->getIntrinsicID(), Ty); SmallVector OpBundles; II->getOperandBundlesAsDefs(OpBundles); CallInst *NewCI = CallInst::Create(Overload, {InnerTrunc}, OpBundles, II->getName()); NewCI->copyFastMathFlags(II); return NewCI; } } } if (Instruction *I = shrinkInsertElt(FPT, Builder)) return I; Value *Src = FPT.getOperand(0); if (isa(Src) || isa(Src)) { auto *FPCast = cast(Src); if (isKnownExactCastIntToFP(*FPCast)) return CastInst::Create(FPCast->getOpcode(), FPCast->getOperand(0), Ty); } return nullptr; } Instruction *InstCombinerImpl::visitFPExt(CastInst &FPExt) { // If the source operand is a cast from integer to FP and known exact, then // cast the integer operand directly to the destination type. Type *Ty = FPExt.getType(); Value *Src = FPExt.getOperand(0); if (isa(Src) || isa(Src)) { auto *FPCast = cast(Src); if (isKnownExactCastIntToFP(*FPCast)) return CastInst::Create(FPCast->getOpcode(), FPCast->getOperand(0), Ty); } return commonCastTransforms(FPExt); } /// fpto{s/u}i({u/s}itofp(X)) --> X or zext(X) or sext(X) or trunc(X) /// This is safe if the intermediate type has enough bits in its mantissa to /// accurately represent all values of X. For example, this won't work with /// i64 -> float -> i64. Instruction *InstCombinerImpl::foldItoFPtoI(CastInst &FI) { if (!isa(FI.getOperand(0)) && !isa(FI.getOperand(0))) return nullptr; auto *OpI = cast(FI.getOperand(0)); Value *X = OpI->getOperand(0); Type *XType = X->getType(); Type *DestType = FI.getType(); bool IsOutputSigned = isa(FI); // Since we can assume the conversion won't overflow, our decision as to // whether the input will fit in the float should depend on the minimum // of the input range and output range. // This means this is also safe for a signed input and unsigned output, since // a negative input would lead to undefined behavior. if (!isKnownExactCastIntToFP(*OpI)) { // The first cast may not round exactly based on the source integer width // and FP width, but the overflow UB rules can still allow this to fold. // If the destination type is narrow, that means the intermediate FP value // must be large enough to hold the source value exactly. // For example, (uint8_t)((float)(uint32_t 16777217) is undefined behavior. int OutputSize = (int)DestType->getScalarSizeInBits() - IsOutputSigned; if (OutputSize > OpI->getType()->getFPMantissaWidth()) return nullptr; } if (DestType->getScalarSizeInBits() > XType->getScalarSizeInBits()) { bool IsInputSigned = isa(OpI); if (IsInputSigned && IsOutputSigned) return new SExtInst(X, DestType); return new ZExtInst(X, DestType); } if (DestType->getScalarSizeInBits() < XType->getScalarSizeInBits()) return new TruncInst(X, DestType); assert(XType == DestType && "Unexpected types for int to FP to int casts"); return replaceInstUsesWith(FI, X); } Instruction *InstCombinerImpl::visitFPToUI(FPToUIInst &FI) { if (Instruction *I = foldItoFPtoI(FI)) return I; return commonCastTransforms(FI); } Instruction *InstCombinerImpl::visitFPToSI(FPToSIInst &FI) { if (Instruction *I = foldItoFPtoI(FI)) return I; return commonCastTransforms(FI); } Instruction *InstCombinerImpl::visitUIToFP(CastInst &CI) { return commonCastTransforms(CI); } Instruction *InstCombinerImpl::visitSIToFP(CastInst &CI) { return commonCastTransforms(CI); } Instruction *InstCombinerImpl::visitIntToPtr(IntToPtrInst &CI) { // If the source integer type is not the intptr_t type for this target, do a // trunc or zext to the intptr_t type, then inttoptr of it. This allows the // cast to be exposed to other transforms. unsigned AS = CI.getAddressSpace(); if (CI.getOperand(0)->getType()->getScalarSizeInBits() != DL.getPointerSizeInBits(AS)) { Type *Ty = CI.getOperand(0)->getType()->getWithNewType( DL.getIntPtrType(CI.getContext(), AS)); Value *P = Builder.CreateZExtOrTrunc(CI.getOperand(0), Ty); return new IntToPtrInst(P, CI.getType()); } if (Instruction *I = commonCastTransforms(CI)) return I; return nullptr; } /// Implement the transforms for cast of pointer (bitcast/ptrtoint) Instruction *InstCombinerImpl::commonPointerCastTransforms(CastInst &CI) { Value *Src = CI.getOperand(0); if (GetElementPtrInst *GEP = dyn_cast(Src)) { // If casting the result of a getelementptr instruction with no offset, turn // this into a cast of the original pointer! if (GEP->hasAllZeroIndices() && // If CI is an addrspacecast and GEP changes the poiner type, merging // GEP into CI would undo canonicalizing addrspacecast with different // pointer types, causing infinite loops. (!isa(CI) || GEP->getType() == GEP->getPointerOperandType())) { // Changing the cast operand is usually not a good idea but it is safe // here because the pointer operand is being replaced with another // pointer operand so the opcode doesn't need to change. return replaceOperand(CI, 0, GEP->getOperand(0)); } } return commonCastTransforms(CI); } Instruction *InstCombinerImpl::visitPtrToInt(PtrToIntInst &CI) { // If the destination integer type is not the intptr_t type for this target, // do a ptrtoint to intptr_t then do a trunc or zext. This allows the cast // to be exposed to other transforms. Value *SrcOp = CI.getPointerOperand(); Type *SrcTy = SrcOp->getType(); Type *Ty = CI.getType(); unsigned AS = CI.getPointerAddressSpace(); unsigned TySize = Ty->getScalarSizeInBits(); unsigned PtrSize = DL.getPointerSizeInBits(AS); if (TySize != PtrSize) { Type *IntPtrTy = SrcTy->getWithNewType(DL.getIntPtrType(CI.getContext(), AS)); Value *P = Builder.CreatePtrToInt(SrcOp, IntPtrTy); return CastInst::CreateIntegerCast(P, Ty, /*isSigned=*/false); } Value *Vec, *Scalar, *Index; if (match(SrcOp, m_OneUse(m_InsertElt(m_IntToPtr(m_Value(Vec)), m_Value(Scalar), m_Value(Index)))) && Vec->getType() == Ty) { assert(Vec->getType()->getScalarSizeInBits() == PtrSize && "Wrong type"); // Convert the scalar to int followed by insert to eliminate one cast: // p2i (ins (i2p Vec), Scalar, Index --> ins Vec, (p2i Scalar), Index Value *NewCast = Builder.CreatePtrToInt(Scalar, Ty->getScalarType()); return InsertElementInst::Create(Vec, NewCast, Index); } return commonPointerCastTransforms(CI); } /// This input value (which is known to have vector type) is being zero extended /// or truncated to the specified vector type. Since the zext/trunc is done /// using an integer type, we have a (bitcast(cast(bitcast))) pattern, /// endianness will impact which end of the vector that is extended or /// truncated. /// /// A vector is always stored with index 0 at the lowest address, which /// corresponds to the most significant bits for a big endian stored integer and /// the least significant bits for little endian. A trunc/zext of an integer /// impacts the big end of the integer. Thus, we need to add/remove elements at /// the front of the vector for big endian targets, and the back of the vector /// for little endian targets. /// /// Try to replace it with a shuffle (and vector/vector bitcast) if possible. /// /// The source and destination vector types may have different element types. static Instruction * optimizeVectorResizeWithIntegerBitCasts(Value *InVal, VectorType *DestTy, InstCombinerImpl &IC) { // We can only do this optimization if the output is a multiple of the input // element size, or the input is a multiple of the output element size. // Convert the input type to have the same element type as the output. VectorType *SrcTy = cast(InVal->getType()); if (SrcTy->getElementType() != DestTy->getElementType()) { // The input types don't need to be identical, but for now they must be the // same size. There is no specific reason we couldn't handle things like // <4 x i16> -> <4 x i32> by bitcasting to <2 x i32> but haven't gotten // there yet. if (SrcTy->getElementType()->getPrimitiveSizeInBits() != DestTy->getElementType()->getPrimitiveSizeInBits()) return nullptr; SrcTy = FixedVectorType::get(DestTy->getElementType(), cast(SrcTy)->getNumElements()); InVal = IC.Builder.CreateBitCast(InVal, SrcTy); } bool IsBigEndian = IC.getDataLayout().isBigEndian(); unsigned SrcElts = cast(SrcTy)->getNumElements(); unsigned DestElts = cast(DestTy)->getNumElements(); assert(SrcElts != DestElts && "Element counts should be different."); // Now that the element types match, get the shuffle mask and RHS of the // shuffle to use, which depends on whether we're increasing or decreasing the // size of the input. SmallVector ShuffleMaskStorage; ArrayRef ShuffleMask; Value *V2; // Produce an identify shuffle mask for the src vector. ShuffleMaskStorage.resize(SrcElts); std::iota(ShuffleMaskStorage.begin(), ShuffleMaskStorage.end(), 0); if (SrcElts > DestElts) { // If we're shrinking the number of elements (rewriting an integer // truncate), just shuffle in the elements corresponding to the least // significant bits from the input and use undef as the second shuffle // input. V2 = UndefValue::get(SrcTy); // Make sure the shuffle mask selects the "least significant bits" by // keeping elements from back of the src vector for big endian, and from the // front for little endian. ShuffleMask = ShuffleMaskStorage; if (IsBigEndian) ShuffleMask = ShuffleMask.take_back(DestElts); else ShuffleMask = ShuffleMask.take_front(DestElts); } else { // If we're increasing the number of elements (rewriting an integer zext), // shuffle in all of the elements from InVal. Fill the rest of the result // elements with zeros from a constant zero. V2 = Constant::getNullValue(SrcTy); // Use first elt from V2 when indicating zero in the shuffle mask. uint32_t NullElt = SrcElts; // Extend with null values in the "most significant bits" by adding elements // in front of the src vector for big endian, and at the back for little // endian. unsigned DeltaElts = DestElts - SrcElts; if (IsBigEndian) ShuffleMaskStorage.insert(ShuffleMaskStorage.begin(), DeltaElts, NullElt); else ShuffleMaskStorage.append(DeltaElts, NullElt); ShuffleMask = ShuffleMaskStorage; } return new ShuffleVectorInst(InVal, V2, ShuffleMask); } static bool isMultipleOfTypeSize(unsigned Value, Type *Ty) { return Value % Ty->getPrimitiveSizeInBits() == 0; } static unsigned getTypeSizeIndex(unsigned Value, Type *Ty) { return Value / Ty->getPrimitiveSizeInBits(); } /// V is a value which is inserted into a vector of VecEltTy. /// Look through the value to see if we can decompose it into /// insertions into the vector. See the example in the comment for /// OptimizeIntegerToVectorInsertions for the pattern this handles. /// The type of V is always a non-zero multiple of VecEltTy's size. /// Shift is the number of bits between the lsb of V and the lsb of /// the vector. /// /// This returns false if the pattern can't be matched or true if it can, /// filling in Elements with the elements found here. static bool collectInsertionElements(Value *V, unsigned Shift, SmallVectorImpl &Elements, Type *VecEltTy, bool isBigEndian) { assert(isMultipleOfTypeSize(Shift, VecEltTy) && "Shift should be a multiple of the element type size"); // Undef values never contribute useful bits to the result. if (isa(V)) return true; // If we got down to a value of the right type, we win, try inserting into the // right element. if (V->getType() == VecEltTy) { // Inserting null doesn't actually insert any elements. if (Constant *C = dyn_cast(V)) if (C->isNullValue()) return true; unsigned ElementIndex = getTypeSizeIndex(Shift, VecEltTy); if (isBigEndian) ElementIndex = Elements.size() - ElementIndex - 1; // Fail if multiple elements are inserted into this slot. if (Elements[ElementIndex]) return false; Elements[ElementIndex] = V; return true; } if (Constant *C = dyn_cast(V)) { // Figure out the # elements this provides, and bitcast it or slice it up // as required. unsigned NumElts = getTypeSizeIndex(C->getType()->getPrimitiveSizeInBits(), VecEltTy); // If the constant is the size of a vector element, we just need to bitcast // it to the right type so it gets properly inserted. if (NumElts == 1) return collectInsertionElements(ConstantExpr::getBitCast(C, VecEltTy), Shift, Elements, VecEltTy, isBigEndian); // Okay, this is a constant that covers multiple elements. Slice it up into // pieces and insert each element-sized piece into the vector. if (!isa(C->getType())) C = ConstantExpr::getBitCast(C, IntegerType::get(V->getContext(), C->getType()->getPrimitiveSizeInBits())); unsigned ElementSize = VecEltTy->getPrimitiveSizeInBits(); Type *ElementIntTy = IntegerType::get(C->getContext(), ElementSize); for (unsigned i = 0; i != NumElts; ++i) { unsigned ShiftI = Shift+i*ElementSize; Constant *Piece = ConstantExpr::getLShr(C, ConstantInt::get(C->getType(), ShiftI)); Piece = ConstantExpr::getTrunc(Piece, ElementIntTy); if (!collectInsertionElements(Piece, ShiftI, Elements, VecEltTy, isBigEndian)) return false; } return true; } if (!V->hasOneUse()) return false; Instruction *I = dyn_cast(V); if (!I) return false; switch (I->getOpcode()) { default: return false; // Unhandled case. case Instruction::BitCast: return collectInsertionElements(I->getOperand(0), Shift, Elements, VecEltTy, isBigEndian); case Instruction::ZExt: if (!isMultipleOfTypeSize( I->getOperand(0)->getType()->getPrimitiveSizeInBits(), VecEltTy)) return false; return collectInsertionElements(I->getOperand(0), Shift, Elements, VecEltTy, isBigEndian); case Instruction::Or: return collectInsertionElements(I->getOperand(0), Shift, Elements, VecEltTy, isBigEndian) && collectInsertionElements(I->getOperand(1), Shift, Elements, VecEltTy, isBigEndian); case Instruction::Shl: { // Must be shifting by a constant that is a multiple of the element size. ConstantInt *CI = dyn_cast(I->getOperand(1)); if (!CI) return false; Shift += CI->getZExtValue(); if (!isMultipleOfTypeSize(Shift, VecEltTy)) return false; return collectInsertionElements(I->getOperand(0), Shift, Elements, VecEltTy, isBigEndian); } } } /// If the input is an 'or' instruction, we may be doing shifts and ors to /// assemble the elements of the vector manually. /// Try to rip the code out and replace it with insertelements. This is to /// optimize code like this: /// /// %tmp37 = bitcast float %inc to i32 /// %tmp38 = zext i32 %tmp37 to i64 /// %tmp31 = bitcast float %inc5 to i32 /// %tmp32 = zext i32 %tmp31 to i64 /// %tmp33 = shl i64 %tmp32, 32 /// %ins35 = or i64 %tmp33, %tmp38 /// %tmp43 = bitcast i64 %ins35 to <2 x float> /// /// Into two insertelements that do "buildvector{%inc, %inc5}". static Value *optimizeIntegerToVectorInsertions(BitCastInst &CI, InstCombinerImpl &IC) { auto *DestVecTy = cast(CI.getType()); Value *IntInput = CI.getOperand(0); SmallVector Elements(DestVecTy->getNumElements()); if (!collectInsertionElements(IntInput, 0, Elements, DestVecTy->getElementType(), IC.getDataLayout().isBigEndian())) return nullptr; // If we succeeded, we know that all of the element are specified by Elements // or are zero if Elements has a null entry. Recast this as a set of // insertions. Value *Result = Constant::getNullValue(CI.getType()); for (unsigned i = 0, e = Elements.size(); i != e; ++i) { if (!Elements[i]) continue; // Unset element. Result = IC.Builder.CreateInsertElement(Result, Elements[i], IC.Builder.getInt32(i)); } return Result; } /// Canonicalize scalar bitcasts of extracted elements into a bitcast of the /// vector followed by extract element. The backend tends to handle bitcasts of /// vectors better than bitcasts of scalars because vector registers are /// usually not type-specific like scalar integer or scalar floating-point. static Instruction *canonicalizeBitCastExtElt(BitCastInst &BitCast, InstCombinerImpl &IC) { // TODO: Create and use a pattern matcher for ExtractElementInst. auto *ExtElt = dyn_cast(BitCast.getOperand(0)); if (!ExtElt || !ExtElt->hasOneUse()) return nullptr; // The bitcast must be to a vectorizable type, otherwise we can't make a new // type to extract from. Type *DestType = BitCast.getType(); if (!VectorType::isValidElementType(DestType)) return nullptr; auto *NewVecType = VectorType::get(DestType, ExtElt->getVectorOperandType()); auto *NewBC = IC.Builder.CreateBitCast(ExtElt->getVectorOperand(), NewVecType, "bc"); return ExtractElementInst::Create(NewBC, ExtElt->getIndexOperand()); } /// Change the type of a bitwise logic operation if we can eliminate a bitcast. static Instruction *foldBitCastBitwiseLogic(BitCastInst &BitCast, InstCombiner::BuilderTy &Builder) { Type *DestTy = BitCast.getType(); BinaryOperator *BO; if (!DestTy->isIntOrIntVectorTy() || !match(BitCast.getOperand(0), m_OneUse(m_BinOp(BO))) || !BO->isBitwiseLogicOp()) return nullptr; // FIXME: This transform is restricted to vector types to avoid backend // problems caused by creating potentially illegal operations. If a fix-up is // added to handle that situation, we can remove this check. if (!DestTy->isVectorTy() || !BO->getType()->isVectorTy()) return nullptr; Value *X; if (match(BO->getOperand(0), m_OneUse(m_BitCast(m_Value(X)))) && X->getType() == DestTy && !isa(X)) { // bitcast(logic(bitcast(X), Y)) --> logic'(X, bitcast(Y)) Value *CastedOp1 = Builder.CreateBitCast(BO->getOperand(1), DestTy); return BinaryOperator::Create(BO->getOpcode(), X, CastedOp1); } if (match(BO->getOperand(1), m_OneUse(m_BitCast(m_Value(X)))) && X->getType() == DestTy && !isa(X)) { // bitcast(logic(Y, bitcast(X))) --> logic'(bitcast(Y), X) Value *CastedOp0 = Builder.CreateBitCast(BO->getOperand(0), DestTy); return BinaryOperator::Create(BO->getOpcode(), CastedOp0, X); } // Canonicalize vector bitcasts to come before vector bitwise logic with a // constant. This eases recognition of special constants for later ops. // Example: // icmp u/s (a ^ signmask), (b ^ signmask) --> icmp s/u a, b Constant *C; if (match(BO->getOperand(1), m_Constant(C))) { // bitcast (logic X, C) --> logic (bitcast X, C') Value *CastedOp0 = Builder.CreateBitCast(BO->getOperand(0), DestTy); Value *CastedC = Builder.CreateBitCast(C, DestTy); return BinaryOperator::Create(BO->getOpcode(), CastedOp0, CastedC); } return nullptr; } /// Change the type of a select if we can eliminate a bitcast. static Instruction *foldBitCastSelect(BitCastInst &BitCast, InstCombiner::BuilderTy &Builder) { Value *Cond, *TVal, *FVal; if (!match(BitCast.getOperand(0), m_OneUse(m_Select(m_Value(Cond), m_Value(TVal), m_Value(FVal))))) return nullptr; // A vector select must maintain the same number of elements in its operands. Type *CondTy = Cond->getType(); Type *DestTy = BitCast.getType(); if (auto *CondVTy = dyn_cast(CondTy)) if (!DestTy->isVectorTy() || CondVTy->getElementCount() != cast(DestTy)->getElementCount()) return nullptr; // FIXME: This transform is restricted from changing the select between // scalars and vectors to avoid backend problems caused by creating // potentially illegal operations. If a fix-up is added to handle that // situation, we can remove this check. if (DestTy->isVectorTy() != TVal->getType()->isVectorTy()) return nullptr; auto *Sel = cast(BitCast.getOperand(0)); Value *X; if (match(TVal, m_OneUse(m_BitCast(m_Value(X)))) && X->getType() == DestTy && !isa(X)) { // bitcast(select(Cond, bitcast(X), Y)) --> select'(Cond, X, bitcast(Y)) Value *CastedVal = Builder.CreateBitCast(FVal, DestTy); return SelectInst::Create(Cond, X, CastedVal, "", nullptr, Sel); } if (match(FVal, m_OneUse(m_BitCast(m_Value(X)))) && X->getType() == DestTy && !isa(X)) { // bitcast(select(Cond, Y, bitcast(X))) --> select'(Cond, bitcast(Y), X) Value *CastedVal = Builder.CreateBitCast(TVal, DestTy); return SelectInst::Create(Cond, CastedVal, X, "", nullptr, Sel); } return nullptr; } /// Check if all users of CI are StoreInsts. static bool hasStoreUsersOnly(CastInst &CI) { for (User *U : CI.users()) { if (!isa(U)) return false; } return true; } /// This function handles following case /// /// A -> B cast /// PHI /// B -> A cast /// /// All the related PHI nodes can be replaced by new PHI nodes with type A. /// The uses of \p CI can be changed to the new PHI node corresponding to \p PN. Instruction *InstCombinerImpl::optimizeBitCastFromPhi(CastInst &CI, PHINode *PN) { // BitCast used by Store can be handled in InstCombineLoadStoreAlloca.cpp. if (hasStoreUsersOnly(CI)) return nullptr; Value *Src = CI.getOperand(0); Type *SrcTy = Src->getType(); // Type B Type *DestTy = CI.getType(); // Type A SmallVector PhiWorklist; SmallSetVector OldPhiNodes; // Find all of the A->B casts and PHI nodes. // We need to inspect all related PHI nodes, but PHIs can be cyclic, so // OldPhiNodes is used to track all known PHI nodes, before adding a new // PHI to PhiWorklist, it is checked against and added to OldPhiNodes first. PhiWorklist.push_back(PN); OldPhiNodes.insert(PN); while (!PhiWorklist.empty()) { auto *OldPN = PhiWorklist.pop_back_val(); for (Value *IncValue : OldPN->incoming_values()) { if (isa(IncValue)) continue; if (auto *LI = dyn_cast(IncValue)) { // If there is a sequence of one or more load instructions, each loaded // value is used as address of later load instruction, bitcast is // necessary to change the value type, don't optimize it. For // simplicity we give up if the load address comes from another load. Value *Addr = LI->getOperand(0); if (Addr == &CI || isa(Addr)) return nullptr; // Don't tranform "load <256 x i32>, <256 x i32>*" to // "load x86_amx, x86_amx*", because x86_amx* is invalid. // TODO: Remove this check when bitcast between vector and x86_amx // is replaced with a specific intrinsic. if (DestTy->isX86_AMXTy()) return nullptr; if (LI->hasOneUse() && LI->isSimple()) continue; // If a LoadInst has more than one use, changing the type of loaded // value may create another bitcast. return nullptr; } if (auto *PNode = dyn_cast(IncValue)) { if (OldPhiNodes.insert(PNode)) PhiWorklist.push_back(PNode); continue; } auto *BCI = dyn_cast(IncValue); // We can't handle other instructions. if (!BCI) return nullptr; // Verify it's a A->B cast. Type *TyA = BCI->getOperand(0)->getType(); Type *TyB = BCI->getType(); if (TyA != DestTy || TyB != SrcTy) return nullptr; } } // Check that each user of each old PHI node is something that we can // rewrite, so that all of the old PHI nodes can be cleaned up afterwards. for (auto *OldPN : OldPhiNodes) { for (User *V : OldPN->users()) { if (auto *SI = dyn_cast(V)) { if (!SI->isSimple() || SI->getOperand(0) != OldPN) return nullptr; } else if (auto *BCI = dyn_cast(V)) { // Verify it's a B->A cast. Type *TyB = BCI->getOperand(0)->getType(); Type *TyA = BCI->getType(); if (TyA != DestTy || TyB != SrcTy) return nullptr; } else if (auto *PHI = dyn_cast(V)) { // As long as the user is another old PHI node, then even if we don't // rewrite it, the PHI web we're considering won't have any users // outside itself, so it'll be dead. if (OldPhiNodes.count(PHI) == 0) return nullptr; } else { return nullptr; } } } // For each old PHI node, create a corresponding new PHI node with a type A. SmallDenseMap NewPNodes; for (auto *OldPN : OldPhiNodes) { Builder.SetInsertPoint(OldPN); PHINode *NewPN = Builder.CreatePHI(DestTy, OldPN->getNumOperands()); NewPNodes[OldPN] = NewPN; } // Fill in the operands of new PHI nodes. for (auto *OldPN : OldPhiNodes) { PHINode *NewPN = NewPNodes[OldPN]; for (unsigned j = 0, e = OldPN->getNumOperands(); j != e; ++j) { Value *V = OldPN->getOperand(j); Value *NewV = nullptr; if (auto *C = dyn_cast(V)) { NewV = ConstantExpr::getBitCast(C, DestTy); } else if (auto *LI = dyn_cast(V)) { // Explicitly perform load combine to make sure no opposing transform // can remove the bitcast in the meantime and trigger an infinite loop. Builder.SetInsertPoint(LI); NewV = combineLoadToNewType(*LI, DestTy); // Remove the old load and its use in the old phi, which itself becomes // dead once the whole transform finishes. replaceInstUsesWith(*LI, PoisonValue::get(LI->getType())); eraseInstFromFunction(*LI); } else if (auto *BCI = dyn_cast(V)) { NewV = BCI->getOperand(0); } else if (auto *PrevPN = dyn_cast(V)) { NewV = NewPNodes[PrevPN]; } assert(NewV); NewPN->addIncoming(NewV, OldPN->getIncomingBlock(j)); } } // Traverse all accumulated PHI nodes and process its users, // which are Stores and BitcCasts. Without this processing // NewPHI nodes could be replicated and could lead to extra // moves generated after DeSSA. // If there is a store with type B, change it to type A. // Replace users of BitCast B->A with NewPHI. These will help // later to get rid off a closure formed by OldPHI nodes. Instruction *RetVal = nullptr; for (auto *OldPN : OldPhiNodes) { PHINode *NewPN = NewPNodes[OldPN]; for (User *V : make_early_inc_range(OldPN->users())) { if (auto *SI = dyn_cast(V)) { assert(SI->isSimple() && SI->getOperand(0) == OldPN); Builder.SetInsertPoint(SI); auto *NewBC = cast(Builder.CreateBitCast(NewPN, SrcTy)); SI->setOperand(0, NewBC); Worklist.push(SI); assert(hasStoreUsersOnly(*NewBC)); } else if (auto *BCI = dyn_cast(V)) { Type *TyB = BCI->getOperand(0)->getType(); Type *TyA = BCI->getType(); assert(TyA == DestTy && TyB == SrcTy); (void) TyA; (void) TyB; Instruction *I = replaceInstUsesWith(*BCI, NewPN); if (BCI == &CI) RetVal = I; } else if (auto *PHI = dyn_cast(V)) { assert(OldPhiNodes.contains(PHI)); (void) PHI; } else { llvm_unreachable("all uses should be handled"); } } } return RetVal; } static Instruction *convertBitCastToGEP(BitCastInst &CI, IRBuilderBase &Builder, const DataLayout &DL) { Value *Src = CI.getOperand(0); PointerType *SrcPTy = cast(Src->getType()); PointerType *DstPTy = cast(CI.getType()); // Bitcasts involving opaque pointers cannot be converted into a GEP. if (SrcPTy->isOpaque() || DstPTy->isOpaque()) return nullptr; Type *DstElTy = DstPTy->getElementType(); Type *SrcElTy = SrcPTy->getElementType(); // When the type pointed to is not sized the cast cannot be // turned into a gep. if (!SrcElTy->isSized()) return nullptr; // If the source and destination are pointers, and this cast is equivalent // to a getelementptr X, 0, 0, 0... turn it into the appropriate gep. // This can enhance SROA and other transforms that want type-safe pointers. unsigned NumZeros = 0; while (SrcElTy && SrcElTy != DstElTy) { SrcElTy = GetElementPtrInst::getTypeAtIndex(SrcElTy, (uint64_t)0); ++NumZeros; } // If we found a path from the src to dest, create the getelementptr now. if (SrcElTy == DstElTy) { SmallVector Idxs(NumZeros + 1, Builder.getInt32(0)); GetElementPtrInst *GEP = GetElementPtrInst::Create(SrcPTy->getElementType(), Src, Idxs); // If the source pointer is dereferenceable, then assume it points to an // allocated object and apply "inbounds" to the GEP. bool CanBeNull, CanBeFreed; if (Src->getPointerDereferenceableBytes(DL, CanBeNull, CanBeFreed)) { // In a non-default address space (not 0), a null pointer can not be // assumed inbounds, so ignore that case (dereferenceable_or_null). // The reason is that 'null' is not treated differently in these address // spaces, and we consequently ignore the 'gep inbounds' special case // for 'null' which allows 'inbounds' on 'null' if the indices are // zeros. if (SrcPTy->getAddressSpace() == 0 || !CanBeNull) GEP->setIsInBounds(); } return GEP; } return nullptr; } Instruction *InstCombinerImpl::visitBitCast(BitCastInst &CI) { // If the operands are integer typed then apply the integer transforms, // otherwise just apply the common ones. Value *Src = CI.getOperand(0); Type *SrcTy = Src->getType(); Type *DestTy = CI.getType(); // Get rid of casts from one type to the same type. These are useless and can // be replaced by the operand. if (DestTy == Src->getType()) return replaceInstUsesWith(CI, Src); if (isa(SrcTy) && isa(DestTy)) { // If we are casting a alloca to a pointer to a type of the same // size, rewrite the allocation instruction to allocate the "right" type. // There is no need to modify malloc calls because it is their bitcast that // needs to be cleaned up. if (AllocaInst *AI = dyn_cast(Src)) if (Instruction *V = PromoteCastOfAllocation(CI, *AI)) return V; if (Instruction *I = convertBitCastToGEP(CI, Builder, DL)) return I; } if (FixedVectorType *DestVTy = dyn_cast(DestTy)) { // Beware: messing with this target-specific oddity may cause trouble. if (DestVTy->getNumElements() == 1 && SrcTy->isX86_MMXTy()) { Value *Elem = Builder.CreateBitCast(Src, DestVTy->getElementType()); return InsertElementInst::Create(PoisonValue::get(DestTy), Elem, Constant::getNullValue(Type::getInt32Ty(CI.getContext()))); } if (isa(SrcTy)) { // If this is a cast from an integer to vector, check to see if the input // is a trunc or zext of a bitcast from vector. If so, we can replace all // the casts with a shuffle and (potentially) a bitcast. if (isa(Src) || isa(Src)) { CastInst *SrcCast = cast(Src); if (BitCastInst *BCIn = dyn_cast(SrcCast->getOperand(0))) if (isa(BCIn->getOperand(0)->getType())) if (Instruction *I = optimizeVectorResizeWithIntegerBitCasts( BCIn->getOperand(0), cast(DestTy), *this)) return I; } // If the input is an 'or' instruction, we may be doing shifts and ors to // assemble the elements of the vector manually. Try to rip the code out // and replace it with insertelements. if (Value *V = optimizeIntegerToVectorInsertions(CI, *this)) return replaceInstUsesWith(CI, V); } } if (FixedVectorType *SrcVTy = dyn_cast(SrcTy)) { if (SrcVTy->getNumElements() == 1) { // If our destination is not a vector, then make this a straight // scalar-scalar cast. if (!DestTy->isVectorTy()) { Value *Elem = Builder.CreateExtractElement(Src, Constant::getNullValue(Type::getInt32Ty(CI.getContext()))); return CastInst::Create(Instruction::BitCast, Elem, DestTy); } // Otherwise, see if our source is an insert. If so, then use the scalar // component directly: // bitcast (inselt <1 x elt> V, X, 0) to --> bitcast X to if (auto *InsElt = dyn_cast(Src)) return new BitCastInst(InsElt->getOperand(1), DestTy); } } if (auto *Shuf = dyn_cast(Src)) { // Okay, we have (bitcast (shuffle ..)). Check to see if this is // a bitcast to a vector with the same # elts. Value *ShufOp0 = Shuf->getOperand(0); Value *ShufOp1 = Shuf->getOperand(1); auto ShufElts = cast(Shuf->getType())->getElementCount(); auto SrcVecElts = cast(ShufOp0->getType())->getElementCount(); if (Shuf->hasOneUse() && DestTy->isVectorTy() && cast(DestTy)->getElementCount() == ShufElts && ShufElts == SrcVecElts) { BitCastInst *Tmp; // If either of the operands is a cast from CI.getType(), then // evaluating the shuffle in the casted destination's type will allow // us to eliminate at least one cast. if (((Tmp = dyn_cast(ShufOp0)) && Tmp->getOperand(0)->getType() == DestTy) || ((Tmp = dyn_cast(ShufOp1)) && Tmp->getOperand(0)->getType() == DestTy)) { Value *LHS = Builder.CreateBitCast(ShufOp0, DestTy); Value *RHS = Builder.CreateBitCast(ShufOp1, DestTy); // Return a new shuffle vector. Use the same element ID's, as we // know the vector types match #elts. return new ShuffleVectorInst(LHS, RHS, Shuf->getShuffleMask()); } } // A bitcasted-to-scalar and byte-reversing shuffle is better recognized as // a byte-swap: // bitcast (shuf X, undef, ) --> bswap (bitcast X) // TODO: We should match the related pattern for bitreverse. if (DestTy->isIntegerTy() && DL.isLegalInteger(DestTy->getScalarSizeInBits()) && SrcTy->getScalarSizeInBits() == 8 && ShufElts.getKnownMinValue() % 2 == 0 && Shuf->hasOneUse() && Shuf->isReverse()) { assert(ShufOp0->getType() == SrcTy && "Unexpected shuffle mask"); assert(match(ShufOp1, m_Undef()) && "Unexpected shuffle op"); Function *Bswap = Intrinsic::getDeclaration(CI.getModule(), Intrinsic::bswap, DestTy); Value *ScalarX = Builder.CreateBitCast(ShufOp0, DestTy); return CallInst::Create(Bswap, { ScalarX }); } } // Handle the A->B->A cast, and there is an intervening PHI node. if (PHINode *PN = dyn_cast(Src)) if (Instruction *I = optimizeBitCastFromPhi(CI, PN)) return I; if (Instruction *I = canonicalizeBitCastExtElt(CI, *this)) return I; if (Instruction *I = foldBitCastBitwiseLogic(CI, Builder)) return I; if (Instruction *I = foldBitCastSelect(CI, Builder)) return I; if (SrcTy->isPointerTy()) return commonPointerCastTransforms(CI); return commonCastTransforms(CI); } Instruction *InstCombinerImpl::visitAddrSpaceCast(AddrSpaceCastInst &CI) { // If the destination pointer element type is not the same as the source's // first do a bitcast to the destination type, and then the addrspacecast. // This allows the cast to be exposed to other transforms. Value *Src = CI.getOperand(0); PointerType *SrcTy = cast(Src->getType()->getScalarType()); PointerType *DestTy = cast(CI.getType()->getScalarType()); if (!SrcTy->hasSameElementTypeAs(DestTy)) { Type *MidTy = PointerType::getWithSamePointeeType(DestTy, SrcTy->getAddressSpace()); // Handle vectors of pointers. if (VectorType *VT = dyn_cast(CI.getType())) MidTy = VectorType::get(MidTy, VT->getElementCount()); Value *NewBitCast = Builder.CreateBitCast(Src, MidTy); return new AddrSpaceCastInst(NewBitCast, CI.getType()); } return commonPointerCastTransforms(CI); }