diff --git a/llvm/lib/Transforms/InstCombine/CMakeLists.txt b/llvm/lib/Transforms/InstCombine/CMakeLists.txt --- a/llvm/lib/Transforms/InstCombine/CMakeLists.txt +++ b/llvm/lib/Transforms/InstCombine/CMakeLists.txt @@ -12,6 +12,7 @@ InstCombineCompares.cpp InstCombineLoadStoreAlloca.cpp InstCombineMulDivRem.cpp + InstCombineNegator.cpp InstCombinePHI.cpp InstCombineSelect.cpp InstCombineShifts.cpp diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp --- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp @@ -1682,12 +1682,10 @@ if (Instruction *X = foldVectorBinop(I)) return X; - // (A*B)-(A*C) -> A*(B-C) etc - if (Value *V = SimplifyUsingDistributiveLaws(I)) - return replaceInstUsesWith(I, V); + Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1); // If this is a 'B = x-(-A)', change to B = x+A. - Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1); + // We deal with this without involving Negator to preserve NSW flag. if (Value *V = dyn_castNegVal(Op1)) { BinaryOperator *Res = BinaryOperator::CreateAdd(Op0, V); @@ -1704,6 +1702,45 @@ return Res; } + auto TryToNarrowDeduceFlags = [this, &I, &Op0, &Op1]() -> Instruction * { + if (Instruction *Ext = narrowMathIfNoOverflow(I)) + return Ext; + + bool Changed = false; + if (!I.hasNoSignedWrap() && willNotOverflowSignedSub(Op0, Op1, I)) { + Changed = true; + I.setHasNoSignedWrap(true); + } + if (!I.hasNoUnsignedWrap() && willNotOverflowUnsignedSub(Op0, Op1, I)) { + Changed = true; + I.setHasNoUnsignedWrap(true); + } + + return Changed ? &I : nullptr; + }; + + // First, let's try to interpret `sub a, b` as `add a, (sub 0, b)`, + // and let's try to sink `(sub 0, b)` into `b` itself. But only if this isn't + // a pure negation used by a select that looks like abs/nabs. + bool IsNegation = match(Op0, m_ZeroInt()); + if (!IsNegation || none_of(I.users(), [&I, Op1](const User *U) { + const Instruction *UI = dyn_cast(U); + if (!UI) + return false; + return match(UI, + m_Select(m_Value(), m_Specific(Op1), m_Specific(&I))) || + match(UI, m_Select(m_Value(), m_Specific(&I), m_Specific(Op1))); + })) { + if (Value *NegOp1 = Negator::Negate(IsNegation, Op1, *this)) + return BinaryOperator::CreateAdd(NegOp1, Op0); + } + if (IsNegation) + return TryToNarrowDeduceFlags(); // Should have been handled in Negator! + + // (A*B)-(A*C) -> A*(B-C) etc + if (Value *V = SimplifyUsingDistributiveLaws(I)) + return replaceInstUsesWith(I, V); + if (I.getType()->isIntOrIntVectorTy(1)) return BinaryOperator::CreateXor(Op0, Op1); @@ -1720,22 +1757,7 @@ if (match(Op0, m_OneUse(m_Add(m_Value(X), m_AllOnes())))) return BinaryOperator::CreateAdd(Builder.CreateNot(Op1), X); - // Y - (X + 1) --> ~X + Y - if (match(Op1, m_OneUse(m_Add(m_Value(X), m_One())))) - return BinaryOperator::CreateAdd(Builder.CreateNot(X), Op0); - - // Y - ~X --> (X + 1) + Y - if (match(Op1, m_OneUse(m_Not(m_Value(X))))) { - return BinaryOperator::CreateAdd( - Builder.CreateAdd(Op0, ConstantInt::get(I.getType(), 1)), X); - } - if (Constant *C = dyn_cast(Op0)) { - // -f(x) -> f(-x) if possible. - if (match(C, m_Zero())) - if (Value *Neg = freelyNegateValue(Op1)) - return replaceInstUsesWith(I, Neg); - Value *X; if (match(Op1, m_ZExt(m_Value(X))) && X->getType()->isIntOrIntVectorTy(1)) // C - (zext bool) --> bool ? C - 1 : C @@ -1770,26 +1792,12 @@ } const APInt *Op0C; - if (match(Op0, m_APInt(Op0C))) { - if (Op0C->isNullValue() && Op1->hasOneUse()) { - Value *LHS, *RHS; - SelectPatternFlavor SPF = matchSelectPattern(Op1, LHS, RHS).Flavor; - if (SPF == SPF_ABS || SPF == SPF_NABS) { - // This is a negate of an ABS/NABS pattern. Just swap the operands - // of the select. - cast(Op1)->swapValues(); - // Don't swap prof metadata, we didn't change the branch behavior. - return replaceInstUsesWith(I, Op1); - } - } - + if (match(Op0, m_APInt(Op0C)) && Op0C->isMask()) { // Turn this into a xor if LHS is 2^n-1 and the remaining bits are known // zero. - if (Op0C->isMask()) { - KnownBits RHSKnown = computeKnownBits(Op1, 0, &I); - if ((*Op0C | RHSKnown.Zero).isAllOnesValue()) - return BinaryOperator::CreateXor(Op1, Op0); - } + KnownBits RHSKnown = computeKnownBits(Op1, 0, &I); + if ((*Op0C | RHSKnown.Zero).isAllOnesValue()) + return BinaryOperator::CreateXor(Op1, Op0); } { @@ -1919,49 +1927,6 @@ return BinaryOperator::CreateAnd( Op0, Builder.CreateNot(Y, Y->getName() + ".not")); - if (Op1->hasOneUse()) { - Value *Y = nullptr, *Z = nullptr; - Constant *C = nullptr; - - // (X - (Y - Z)) --> (X + (Z - Y)). - if (match(Op1, m_Sub(m_Value(Y), m_Value(Z)))) - return BinaryOperator::CreateAdd(Op0, - Builder.CreateSub(Z, Y, Op1->getName())); - - // Subtracting -1/0 is the same as adding 1/0: - // sub [nsw] Op0, sext(bool Y) -> add [nsw] Op0, zext(bool Y) - // 'nuw' is dropped in favor of the canonical form. - if (match(Op1, m_SExt(m_Value(Y))) && - Y->getType()->getScalarSizeInBits() == 1) { - Value *Zext = Builder.CreateZExt(Y, I.getType()); - BinaryOperator *Add = BinaryOperator::CreateAdd(Op0, Zext); - Add->setHasNoSignedWrap(I.hasNoSignedWrap()); - return Add; - } - // sub [nsw] X, zext(bool Y) -> add [nsw] X, sext(bool Y) - // 'nuw' is dropped in favor of the canonical form. - if (match(Op1, m_ZExt(m_Value(Y))) && Y->getType()->isIntOrIntVectorTy(1)) { - Value *Sext = Builder.CreateSExt(Y, I.getType()); - BinaryOperator *Add = BinaryOperator::CreateAdd(Op0, Sext); - Add->setHasNoSignedWrap(I.hasNoSignedWrap()); - return Add; - } - - // X - A*-B -> X + A*B - // X - -A*B -> X + A*B - Value *A, *B; - if (match(Op1, m_c_Mul(m_Value(A), m_Neg(m_Value(B))))) - return BinaryOperator::CreateAdd(Op0, Builder.CreateMul(A, B)); - - // X - A*C -> X + A*-C - // No need to handle commuted multiply because multiply handling will - // ensure constant will be move to the right hand side. - if (match(Op1, m_Mul(m_Value(A), m_Constant(C))) && !isa(C)) { - Value *NewMul = Builder.CreateMul(A, ConstantExpr::getNeg(C)); - return BinaryOperator::CreateAdd(Op0, NewMul); - } - } - { // ~A - Min/Max(~A, O) -> Max/Min(A, ~O) - A // ~A - Min/Max(O, ~A) -> Max/Min(A, ~O) - A @@ -2036,20 +2001,7 @@ canonicalizeCondSignextOfHighBitExtractToSignextHighBitExtract(I)) return V; - if (Instruction *Ext = narrowMathIfNoOverflow(I)) - return Ext; - - bool Changed = false; - if (!I.hasNoSignedWrap() && willNotOverflowSignedSub(Op0, Op1, I)) { - Changed = true; - I.setHasNoSignedWrap(true); - } - if (!I.hasNoUnsignedWrap() && willNotOverflowUnsignedSub(Op0, Op1, I)) { - Changed = true; - I.setHasNoUnsignedWrap(true); - } - - return Changed ? &I : nullptr; + return TryToNarrowDeduceFlags(); } /// This eliminates floating-point negation in either 'fneg(X)' or diff --git a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h --- a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h +++ b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h @@ -16,6 +16,7 @@ #define LLVM_LIB_TRANSFORMS_INSTCOMBINE_INSTCOMBINEINTERNAL_H #include "llvm/ADT/ArrayRef.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/AliasAnalysis.h" #include "llvm/Analysis/InstructionSimplify.h" #include "llvm/Analysis/TargetFolder.h" @@ -471,7 +472,6 @@ bool shouldChangeType(unsigned FromBitWidth, unsigned ToBitWidth) const; bool shouldChangeType(Type *From, Type *To) const; Value *dyn_castNegVal(Value *V) const; - Value *freelyNegateValue(Value *V); Type *FindElementAtOffset(PointerType *PtrTy, int64_t Offset, SmallVectorImpl &NewIndices); @@ -513,7 +513,7 @@ Instruction *simplifyMaskedStore(IntrinsicInst &II); Instruction *simplifyMaskedGather(IntrinsicInst &II); Instruction *simplifyMaskedScatter(IntrinsicInst &II); - + /// Transform (zext icmp) to bitwise / integer operations in order to /// eliminate it. /// @@ -1014,6 +1014,55 @@ Value *Descale(Value *Val, APInt Scale, bool &NoSignedWrap); }; +namespace { + +// As a default, let's assume that we want to be aggressive, +// and attempt to traverse with no limits in attempt to sink negation. +static constexpr unsigned NegatorDefaultMaxDepth = ~0U; + +// Let's guesstimate that most often we will end up visiting/producing +// fairly small number of new instructions. +static constexpr unsigned NegatorMaxNodesSSO = 16; + +} // namespace + +class Negator final { + /// Top-to-bottom, def-to-use negated instruction tree we produced. + SmallVector NewInstructions; + + using BuilderTy = IRBuilder; + BuilderTy Builder; + + const bool IsTrulyNegation; + + Negator(LLVMContext &C, const DataLayout &DL, bool IsTrulyNegation); + +#if LLVM_ENABLE_STATS + unsigned NumValuesVisitedInThisNegator = 0; + ~Negator(); +#endif + + using Result = std::pair /*NewInstructions*/, + Value * /*NegatedRoot*/>; + + LLVM_NODISCARD Value *visit(Value *V, unsigned Depth); + + /// Recurse depth-first and attempt to sink the negation. + /// FIXME: use worklist? + LLVM_NODISCARD Optional run(Value *Root); + + Negator(const Negator &) = delete; + Negator(Negator &&) = delete; + Negator &operator=(const Negator &) = delete; + Negator &operator=(Negator &&) = delete; + +public: + /// Attempt to negate \p Root. Retuns nullptr if negation can't be performed, + /// otherwise returns negated value. + LLVM_NODISCARD static Value *Negate(bool LHSIsZero, Value *Root, + InstCombiner &IC); +}; + } // end namespace llvm #undef DEBUG_TYPE diff --git a/llvm/lib/Transforms/InstCombine/InstCombineNegator.cpp b/llvm/lib/Transforms/InstCombine/InstCombineNegator.cpp new file mode 100644 --- /dev/null +++ b/llvm/lib/Transforms/InstCombine/InstCombineNegator.cpp @@ -0,0 +1,377 @@ +//===- InstCombineNegator.cpp -----------------------------------*- C++ -*-===// +// +// 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 sinking of negation into expression trees, +// as long as that can be done without increasing instruction count. +// +//===----------------------------------------------------------------------===// + +#include "InstCombineInternal.h" +#include "llvm/ADT/APInt.h" +#include "llvm/ADT/ArrayRef.h" +#include "llvm/ADT/None.h" +#include "llvm/ADT/Optional.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/ADT/StringRef.h" +#include "llvm/ADT/Twine.h" +#include "llvm/ADT/iterator_range.h" +#include "llvm/Analysis/TargetFolder.h" +#include "llvm/Analysis/ValueTracking.h" +#include "llvm/IR/Constant.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DebugLoc.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/IRBuilder.h" +#include "llvm/IR/Instruction.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/PatternMatch.h" +#include "llvm/IR/Type.h" +#include "llvm/IR/Use.h" +#include "llvm/IR/User.h" +#include "llvm/IR/Value.h" +#include "llvm/Support/Casting.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Compiler.h" +#include "llvm/Support/DebugCounter.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/raw_ostream.h" +#include +#include +#include + +using namespace llvm; + +#define DEBUG_TYPE "instcombine" + +STATISTIC(NegatorTotalNegationsAttempted, + "Negator: Number of negations attempted to be sinked"); +STATISTIC(NegatorNumTreesNegated, + "Negator: Number of negations successfully sinked"); +STATISTIC(NegatorMaxDepthVisited, "Negator: Maximal traversal depth ever " + "reached while attempting to sink negation"); +STATISTIC(NegatorTimesDepthLimitReached, + "Negator: How many times did the traversal depth limit was reached " + "during sinking"); +STATISTIC( + NegatorNumValuesVisited, + "Negator: Total number of values visited during attempts to sink negation"); +STATISTIC(NegatorMaxTotalValuesVisited, + "Negator: Maximal number of values ever visited while attempting to " + "sink negation"); +STATISTIC(NegatorNumInstructionsCreatedTotal, + "Negator: Number of new negated instructions created, total"); +STATISTIC(NegatorMaxInstructionsCreated, + "Negator: Maximal number of new instructions created during negation " + "attempt"); +STATISTIC(NegatorNumInstructionsNegatedSuccess, + "Negator: Number of new negated instructions created in successful " + "negation sinking attempts"); + +DEBUG_COUNTER(NegatorCounter, "instcombine-negator", + "Controls Negator transformations in InstCombine pass"); + +static cl::opt + NegatorEnabled("instcombine-negator-enabled", cl::init(true), + cl::desc("Should we attempt to sink negations?")); + +static cl::opt + NegatorMaxDepth("instcombine-negator-max-depth", + cl::init(NegatorDefaultMaxDepth), + cl::desc("What is the maximal lookup depth when trying to " + "check for viability of negation sinking.")); + +Negator::Negator(LLVMContext &C, const DataLayout &DL, bool IsTrulyNegation_) + : Builder(C, TargetFolder(DL), + IRBuilderCallbackInserter([&](Instruction *I) { + ++NegatorNumInstructionsCreatedTotal; + NewInstructions.push_back(I); + })), + IsTrulyNegation(IsTrulyNegation_) {} + +#if LLVM_ENABLE_STATS +Negator::~Negator() { + NegatorMaxTotalValuesVisited.updateMax(NumValuesVisitedInThisNegator); +} +#endif + +// FIXME: can this be reworked into a worklist-based algorithm while preserving +// the depth-first, early bailout traversal? +LLVM_NODISCARD Value *Negator::visit(Value *V, unsigned Depth) { + NegatorMaxDepthVisited.updateMax(Depth); + ++NegatorNumValuesVisited; + +#if LLVM_ENABLE_STATS + ++NumValuesVisitedInThisNegator; +#endif + + // In i1, negation can simply be ignored. + if (V->getType()->isIntOrIntVectorTy(1)) + return V; + + Value *X; + + // -(-(X)) -> X. + if (match(V, m_Neg(m_Value(X)))) + return X; + + // Integral constants can be freely negated. + if (match(V, m_AnyIntegralConstant())) + return ConstantExpr::getNeg(cast(V), /*HasNUW=*/false, + /*HasNSW=*/false); + + // If we have a non-instruction, then give up. + if (!isa(V)) + return nullptr; + + // If we have started with a true negation (i.e. `sub 0, %y`), then if we've + // got instruction that does not require recursive reasoning, we can still + // negate it even if it has other uses, without increasing instruction count. + if (!V->hasOneUse() && !IsTrulyNegation) + return nullptr; + + auto *I = cast(V); + unsigned BitWidth = I->getType()->getScalarSizeInBits(); + + // We must preserve the insertion point and debug info that is set in the + // builder at the time this function is called. + InstCombiner::BuilderTy::InsertPointGuard Guard(Builder); + // And since we are trying to negate instruction I, that tells us about the + // insertion point and the debug info that we need to keep. + Builder.SetInsertPoint(I); + + // In some cases we can give the answer without further recursion. + switch (I->getOpcode()) { + case Instruction::Sub: + // `sub` is always negatible. + return Builder.CreateSub(I->getOperand(1), I->getOperand(0), + I->getName() + ".neg"); + case Instruction::Add: + // `inc` is always negatible. + if (match(I->getOperand(1), m_One())) + return Builder.CreateNot(I->getOperand(0), I->getName() + ".neg"); + break; + case Instruction::Xor: + // `not` is always negatible. + if (match(I, m_Not(m_Value(X)))) + return Builder.CreateAdd(X, ConstantInt::get(X->getType(), 1), + I->getName() + ".neg"); + break; + case Instruction::AShr: + case Instruction::LShr: { + // Right-shift sign bit smear is negatible. + const APInt *Op1Val; + if (match(I->getOperand(1), m_APInt(Op1Val)) && *Op1Val == BitWidth - 1) { + Value *BO = I->getOpcode() == Instruction::AShr + ? Builder.CreateLShr(I->getOperand(0), I->getOperand(1)) + : Builder.CreateAShr(I->getOperand(0), I->getOperand(1)); + if (auto *NewInstr = dyn_cast(BO)) { + NewInstr->copyIRFlags(I); + NewInstr->setName(I->getName() + ".neg"); + } + return BO; + } + break; + } + case Instruction::SDiv: + // `sdiv` is negatible if divisor is not undef/INT_MIN/1. + // While this is normally not behind a use-check, + // let's consider division to be special since it's costly. + if (!I->hasOneUse()) + break; + if (auto *Op1C = dyn_cast(I->getOperand(1))) { + if (!Op1C->containsUndefElement() && Op1C->isNotMinSignedValue() && + Op1C->isNotOneValue()) { + Value *BO = + Builder.CreateSDiv(I->getOperand(0), ConstantExpr::getNeg(Op1C), + I->getName() + ".neg"); + if (auto *NewInstr = dyn_cast(BO)) + NewInstr->setIsExact(I->isExact()); + return BO; + } + } + break; + case Instruction::SExt: + case Instruction::ZExt: + // `*ext` of i1 is always negatible + if (I->getOperand(0)->getType()->isIntOrIntVectorTy(1)) + return I->getOpcode() == Instruction::SExt + ? Builder.CreateZExt(I->getOperand(0), I->getType(), + I->getName() + ".neg") + : Builder.CreateSExt(I->getOperand(0), I->getType(), + I->getName() + ".neg"); + break; + default: + break; // Other instructions require recursive reasoning. + } + + // Rest of the logic is recursive, and if either the current instruction + // has other uses or if it's time to give up then it's time. + if (!V->hasOneUse()) + return nullptr; + if (Depth > NegatorMaxDepth) { + LLVM_DEBUG(dbgs() << "Negator: reached maximal allowed traversal depth in " + << *V << ". Giving up.\n"); + ++NegatorTimesDepthLimitReached; + return nullptr; + } + + switch (I->getOpcode()) { + case Instruction::PHI: { + // `phi` is negatible if all the incoming values are negatible. + PHINode *PHI = cast(I); + SmallVector NegatedIncomingValues(PHI->getNumOperands()); + for (auto I : zip(PHI->incoming_values(), NegatedIncomingValues)) { + if (!(std::get<1>(I) = visit(std::get<0>(I), Depth + 1))) // Early return. + return nullptr; + } + // All incoming values are indeed negatible. Create negated PHI node. + PHINode *NegatedPHI = Builder.CreatePHI( + PHI->getType(), PHI->getNumOperands(), PHI->getName() + ".neg"); + for (auto I : zip(NegatedIncomingValues, PHI->blocks())) + NegatedPHI->addIncoming(std::get<0>(I), std::get<1>(I)); + return NegatedPHI; + } + case Instruction::Select: { + { + // `abs`/`nabs` is always negatible. + Value *LHS, *RHS; + SelectPatternFlavor SPF = + matchSelectPattern(I, LHS, RHS, /*CastOp=*/nullptr, Depth).Flavor; + if (SPF == SPF_ABS || SPF == SPF_NABS) { + auto *NewSelect = cast(I->clone()); + // Just swap the operands of the select. + NewSelect->swapValues(); + // Don't swap prof metadata, we didn't change the branch behavior. + NewSelect->setName(I->getName() + ".neg"); + Builder.Insert(NewSelect); + return NewSelect; + } + } + // `select` is negatible if both hands of `select` are negatible. + Value *NegOp1 = visit(I->getOperand(1), Depth + 1); + if (!NegOp1) // Early return. + return nullptr; + Value *NegOp2 = visit(I->getOperand(2), Depth + 1); + if (!NegOp2) + return nullptr; + // Do preserve the metadata! + return Builder.CreateSelect(I->getOperand(0), NegOp1, NegOp2, + I->getName() + ".neg", /*MDFrom=*/I); + } + case Instruction::Trunc: { + // `trunc` is negatible if its operand is negatible. + Value *NegOp = visit(I->getOperand(0), Depth + 1); + if (!NegOp) // Early return. + return nullptr; + return Builder.CreateTrunc(NegOp, I->getType(), I->getName() + ".neg"); + } + case Instruction::Shl: { + // `shl` is negatible if the first operand is negatible. + Value *NegOp0 = visit(I->getOperand(0), Depth + 1); + if (!NegOp0) // Early return. + return nullptr; + return Builder.CreateShl(NegOp0, I->getOperand(1), I->getName() + ".neg"); + } + case Instruction::Add: { + // `add` is negatible if both of its operands are negatible. + Value *NegOp0 = visit(I->getOperand(0), Depth + 1); + if (!NegOp0) // Early return. + return nullptr; + Value *NegOp1 = visit(I->getOperand(1), Depth + 1); + if (!NegOp1) + return nullptr; + return Builder.CreateAdd(NegOp0, NegOp1, I->getName() + ".neg"); + } + case Instruction::Xor: + // `xor` is negatible if one of its operands is invertible. + // FIXME: InstCombineInverter? But how to connect Inverter and Negator? + if (auto *C = dyn_cast(I->getOperand(1))) { + Value *Xor = Builder.CreateXor(I->getOperand(0), ConstantExpr::getNot(C)); + return Builder.CreateAdd(Xor, ConstantInt::get(Xor->getType(), 1), + I->getName() + ".neg"); + } + return nullptr; + case Instruction::Mul: { + // `mul` is negatible if one of its operands is negatible. + Value *NegatedOp, *OtherOp; + // First try the second operand, in case it's a constant it will be best to + // just invert it instead of sinking the `neg` deeper. + if (Value *NegOp1 = visit(I->getOperand(1), Depth + 1)) { + NegatedOp = NegOp1; + OtherOp = I->getOperand(0); + } else if (Value *NegOp0 = visit(I->getOperand(0), Depth + 1)) { + NegatedOp = NegOp0; + OtherOp = I->getOperand(1); + } else + // Can't negate either of them. + return nullptr; + return Builder.CreateMul(NegatedOp, OtherOp, I->getName() + ".neg"); + } + default: + return nullptr; // Don't know, likely not negatible for free. + } + + llvm_unreachable("Can't get here. We always return from switch."); +}; + +LLVM_NODISCARD Optional Negator::run(Value *Root) { + Value *Negated = visit(Root, /*Depth=*/0); + if (!Negated) { + // We must cleanup newly-inserted instructions, to avoid any potential + // endless combine looping. + llvm::for_each(llvm::reverse(NewInstructions), + [&](Instruction *I) { I->eraseFromParent(); }); + return llvm::None; + } + return std::make_pair(ArrayRef(NewInstructions), Negated); +}; + +LLVM_NODISCARD Value *Negator::Negate(bool LHSIsZero, Value *Root, + InstCombiner &IC) { + ++NegatorTotalNegationsAttempted; + LLVM_DEBUG(dbgs() << "Negator: attempting to sink negation into " << *Root + << "\n"); + + if (!NegatorEnabled || !DebugCounter::shouldExecute(NegatorCounter)) + return nullptr; + + Negator N(Root->getContext(), IC.getDataLayout(), LHSIsZero); + Optional Res = N.run(Root); + if (!Res) { // Negation failed. + LLVM_DEBUG(dbgs() << "Negator: failed to sink negation into " << *Root + << "\n"); + return nullptr; + } + + LLVM_DEBUG(dbgs() << "Negator: successfully sunk negation into " << *Root + << "\n NEW: " << *Res->second << "\n"); + ++NegatorNumTreesNegated; + + // We must temporarily unset the 'current' insertion point and DebugLoc of the + // InstCombine's IRBuilder so that it won't interfere with the ones we have + // already specified when producing negated instructions. + InstCombiner::BuilderTy::InsertPointGuard Guard(IC.Builder); + IC.Builder.ClearInsertionPoint(); + IC.Builder.SetCurrentDebugLocation(DebugLoc()); + + // And finally, we must add newly-created instructions into the InstCombine's + // worklist (in a proper order!) so it can attempt to combine them. + LLVM_DEBUG(dbgs() << "Negator: Propagating " << Res->first.size() + << " instrs to InstCombine\n"); + NegatorMaxInstructionsCreated.updateMax(Res->first.size()); + NegatorNumInstructionsNegatedSuccess += Res->first.size(); + + // They are in def-use order, so nothing fancy, just insert them in order. + llvm::for_each(Res->first, + [&](Instruction *I) { IC.Builder.Insert(I, I->getName()); }); + + // And return the new root. + return Res->second; +}; diff --git a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp --- a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp +++ b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp @@ -853,120 +853,6 @@ return nullptr; } -/// Get negated V (that is 0-V) without increasing instruction count, -/// assuming that the original V will become unused. -Value *InstCombiner::freelyNegateValue(Value *V) { - if (Value *NegV = dyn_castNegVal(V)) - return NegV; - - Instruction *I = dyn_cast(V); - if (!I) - return nullptr; - - unsigned BitWidth = I->getType()->getScalarSizeInBits(); - switch (I->getOpcode()) { - // 0-(zext i1 A) => sext i1 A - case Instruction::ZExt: - if (I->getOperand(0)->getType()->isIntOrIntVectorTy(1)) - return Builder.CreateSExtOrBitCast( - I->getOperand(0), I->getType(), I->getName() + ".neg"); - return nullptr; - - // 0-(sext i1 A) => zext i1 A - case Instruction::SExt: - if (I->getOperand(0)->getType()->isIntOrIntVectorTy(1)) - return Builder.CreateZExtOrBitCast( - I->getOperand(0), I->getType(), I->getName() + ".neg"); - return nullptr; - - // 0-(A lshr (BW-1)) => A ashr (BW-1) - case Instruction::LShr: - if (match(I->getOperand(1), m_SpecificInt(BitWidth - 1))) - return Builder.CreateAShr( - I->getOperand(0), I->getOperand(1), - I->getName() + ".neg", cast(I)->isExact()); - return nullptr; - - // 0-(A ashr (BW-1)) => A lshr (BW-1) - case Instruction::AShr: - if (match(I->getOperand(1), m_SpecificInt(BitWidth - 1))) - return Builder.CreateLShr( - I->getOperand(0), I->getOperand(1), - I->getName() + ".neg", cast(I)->isExact()); - return nullptr; - - // Negation is equivalent to bitwise-not + 1. - case Instruction::Xor: { - // Special case for negate of 'not' - replace with increment: - // 0 - (~A) => ((A ^ -1) ^ -1) + 1 => A + 1 - Value *A; - if (match(I, m_Not(m_Value(A)))) - return Builder.CreateAdd(A, ConstantInt::get(A->getType(), 1), - I->getName() + ".neg"); - - // General case xor (not a 'not') requires creating a new xor, so this has a - // one-use limitation: - // 0 - (A ^ C) => ((A ^ C) ^ -1) + 1 => A ^ ~C + 1 - Constant *C; - if (match(I, m_OneUse(m_Xor(m_Value(A), m_Constant(C))))) { - Value *Xor = Builder.CreateXor(A, ConstantExpr::getNot(C)); - return Builder.CreateAdd(Xor, ConstantInt::get(Xor->getType(), 1), - I->getName() + ".neg"); - } - return nullptr; - } - - default: - break; - } - - // TODO: The "sub" pattern below could also be applied without the one-use - // restriction. Not allowing it for now in line with existing behavior. - if (!I->hasOneUse()) - return nullptr; - - switch (I->getOpcode()) { - // 0-(A-B) => B-A - case Instruction::Sub: - return Builder.CreateSub( - I->getOperand(1), I->getOperand(0), I->getName() + ".neg"); - - // 0-(A sdiv C) => A sdiv (0-C) provided the negation doesn't overflow. - case Instruction::SDiv: { - Constant *C = dyn_cast(I->getOperand(1)); - if (C && !C->containsUndefElement() && C->isNotMinSignedValue() && - C->isNotOneValue()) - return Builder.CreateSDiv(I->getOperand(0), ConstantExpr::getNeg(C), - I->getName() + ".neg", cast(I)->isExact()); - return nullptr; - } - - // 0-(A< (0-A)<getOperand(0))) - return Builder.CreateShl(NegA, I->getOperand(1), I->getName() + ".neg"); - return nullptr; - - // 0-(trunc A) => trunc (0-A) - case Instruction::Trunc: - if (Value *NegA = freelyNegateValue(I->getOperand(0))) - return Builder.CreateTrunc(NegA, I->getType(), I->getName() + ".neg"); - return nullptr; - - // 0-(A*B) => (0-A)*B - // 0-(A*B) => A*(0-B) - case Instruction::Mul: - if (Value *NegA = freelyNegateValue(I->getOperand(0))) - return Builder.CreateMul(NegA, I->getOperand(1), V->getName() + ".neg"); - if (Value *NegB = freelyNegateValue(I->getOperand(1))) - return Builder.CreateMul(I->getOperand(0), NegB, V->getName() + ".neg"); - return nullptr; - - default: - return nullptr; - } -} - static Value *foldOperationIntoSelectOperand(Instruction &I, Value *SO, InstCombiner::BuilderTy &Builder) { if (auto *Cast = dyn_cast(&I)) diff --git a/llvm/test/Transforms/InstCombine/and-or-icmps.ll b/llvm/test/Transforms/InstCombine/and-or-icmps.ll --- a/llvm/test/Transforms/InstCombine/and-or-icmps.ll +++ b/llvm/test/Transforms/InstCombine/and-or-icmps.ll @@ -210,12 +210,20 @@ ; CHECK-LABEL: @simplify_before_foldAndOfICmps( ; CHECK-NEXT: [[A8:%.*]] = alloca i16, align 2 ; CHECK-NEXT: [[L7:%.*]] = load i16, i16* [[A8]], align 2 -; CHECK-NEXT: [[C10:%.*]] = icmp ult i16 [[L7]], 2 +; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i16 [[L7]], -1 +; CHECK-NEXT: [[B11:%.*]] = zext i1 [[TMP1]] to i16 +; CHECK-NEXT: [[C10:%.*]] = icmp ugt i16 [[L7]], [[B11]] +; CHECK-NEXT: [[C5:%.*]] = icmp slt i16 [[L7]], 1 +; CHECK-NEXT: [[C11:%.*]] = icmp ne i16 [[L7]], 0 ; CHECK-NEXT: [[C7:%.*]] = icmp slt i16 [[L7]], 0 -; CHECK-NEXT: [[C18:%.*]] = or i1 [[C7]], [[C10]] -; CHECK-NEXT: [[L7_LOBIT:%.*]] = ashr i16 [[L7]], 15 -; CHECK-NEXT: [[TMP1:%.*]] = sext i16 [[L7_LOBIT]] to i64 -; CHECK-NEXT: [[G26:%.*]] = getelementptr i1, i1* null, i64 [[TMP1]] +; CHECK-NEXT: [[B15:%.*]] = xor i1 [[C7]], [[C10]] +; CHECK-NEXT: [[B19:%.*]] = xor i1 [[C11]], [[B15]] +; CHECK-NEXT: [[TMP2:%.*]] = and i1 [[C10]], [[C5]] +; CHECK-NEXT: [[C3:%.*]] = and i1 [[TMP2]], [[B19]] +; CHECK-NEXT: [[TMP3:%.*]] = xor i1 [[C10]], true +; CHECK-NEXT: [[C18:%.*]] = or i1 [[C7]], [[TMP3]] +; CHECK-NEXT: [[TMP4:%.*]] = sext i1 [[C3]] to i64 +; CHECK-NEXT: [[G26:%.*]] = getelementptr i1, i1* null, i64 [[TMP4]] ; CHECK-NEXT: store i16 [[L7]], i16* undef, align 2 ; CHECK-NEXT: store i1 [[C18]], i1* undef, align 1 ; CHECK-NEXT: store i1* [[G26]], i1** undef, align 8 diff --git a/llvm/test/Transforms/InstCombine/fold-sub-of-not-to-inc-of-add.ll b/llvm/test/Transforms/InstCombine/fold-sub-of-not-to-inc-of-add.ll --- a/llvm/test/Transforms/InstCombine/fold-sub-of-not-to-inc-of-add.ll +++ b/llvm/test/Transforms/InstCombine/fold-sub-of-not-to-inc-of-add.ll @@ -13,8 +13,8 @@ define i32 @p0_scalar(i32 %x, i32 %y) { ; CHECK-LABEL: @p0_scalar( -; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[Y:%.*]], 1 -; CHECK-NEXT: [[T1:%.*]] = add i32 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[T0_NEG:%.*]] = add i32 [[X:%.*]], 1 +; CHECK-NEXT: [[T1:%.*]] = add i32 [[T0_NEG]], [[Y:%.*]] ; CHECK-NEXT: ret i32 [[T1]] ; %t0 = xor i32 %x, -1 @@ -28,8 +28,8 @@ define <4 x i32> @p1_vector_splat(<4 x i32> %x, <4 x i32> %y) { ; CHECK-LABEL: @p1_vector_splat( -; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[Y:%.*]], -; CHECK-NEXT: [[T1:%.*]] = add <4 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[T0_NEG:%.*]] = add <4 x i32> [[X:%.*]], +; CHECK-NEXT: [[T1:%.*]] = add <4 x i32> [[T0_NEG]], [[Y:%.*]] ; CHECK-NEXT: ret <4 x i32> [[T1]] ; %t0 = xor <4 x i32> %x, @@ -39,8 +39,8 @@ define <4 x i32> @p2_vector_undef(<4 x i32> %x, <4 x i32> %y) { ; CHECK-LABEL: @p2_vector_undef( -; CHECK-NEXT: [[TMP1:%.*]] = add <4 x i32> [[Y:%.*]], -; CHECK-NEXT: [[T1:%.*]] = add <4 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[T0_NEG:%.*]] = add <4 x i32> [[X:%.*]], +; CHECK-NEXT: [[T1:%.*]] = add <4 x i32> [[T0_NEG]], [[Y:%.*]] ; CHECK-NEXT: ret <4 x i32> [[T1]] ; %t0 = xor <4 x i32> %x, @@ -85,8 +85,8 @@ define i32 @n5_is_not_not(i32 %x, i32 %y) { ; CHECK-LABEL: @n5_is_not_not( -; CHECK-NEXT: [[T0:%.*]] = xor i32 [[X:%.*]], 2147483647 -; CHECK-NEXT: [[T1:%.*]] = sub i32 [[Y:%.*]], [[T0]] +; CHECK-NEXT: [[T0_NEG:%.*]] = add i32 [[X:%.*]], -2147483647 +; CHECK-NEXT: [[T1:%.*]] = add i32 [[T0_NEG]], [[Y:%.*]] ; CHECK-NEXT: ret i32 [[T1]] ; %t0 = xor i32 %x, 2147483647 ; not -1 diff --git a/llvm/test/Transforms/InstCombine/high-bit-signmask-with-trunc.ll b/llvm/test/Transforms/InstCombine/high-bit-signmask-with-trunc.ll --- a/llvm/test/Transforms/InstCombine/high-bit-signmask-with-trunc.ll +++ b/llvm/test/Transforms/InstCombine/high-bit-signmask-with-trunc.ll @@ -3,9 +3,9 @@ define i32 @t0(i64 %x) { ; CHECK-LABEL: @t0( -; CHECK-NEXT: [[TMP1:%.*]] = ashr i64 [[X:%.*]], 63 -; CHECK-NEXT: [[R:%.*]] = trunc i64 [[TMP1]] to i32 -; CHECK-NEXT: ret i32 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc i64 [[T0_NEG]] to i32 +; CHECK-NEXT: ret i32 [[T1_NEG]] ; %t0 = lshr i64 %x, 63 %t1 = trunc i64 %t0 to i32 @@ -14,9 +14,9 @@ } define i32 @t1_exact(i64 %x) { ; CHECK-LABEL: @t1_exact( -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact i64 [[X:%.*]], 63 -; CHECK-NEXT: [[R:%.*]] = trunc i64 [[TMP1]] to i32 -; CHECK-NEXT: ret i32 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr exact i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc i64 [[T0_NEG]] to i32 +; CHECK-NEXT: ret i32 [[T1_NEG]] ; %t0 = lshr exact i64 %x, 63 %t1 = trunc i64 %t0 to i32 @@ -25,9 +25,9 @@ } define i32 @t2(i64 %x) { ; CHECK-LABEL: @t2( -; CHECK-NEXT: [[TMP1:%.*]] = lshr i64 [[X:%.*]], 63 -; CHECK-NEXT: [[R:%.*]] = trunc i64 [[TMP1]] to i32 -; CHECK-NEXT: ret i32 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = lshr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc i64 [[T0_NEG]] to i32 +; CHECK-NEXT: ret i32 [[T1_NEG]] ; %t0 = ashr i64 %x, 63 %t1 = trunc i64 %t0 to i32 @@ -36,9 +36,9 @@ } define i32 @t3_exact(i64 %x) { ; CHECK-LABEL: @t3_exact( -; CHECK-NEXT: [[TMP1:%.*]] = lshr exact i64 [[X:%.*]], 63 -; CHECK-NEXT: [[R:%.*]] = trunc i64 [[TMP1]] to i32 -; CHECK-NEXT: ret i32 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = lshr exact i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc i64 [[T0_NEG]] to i32 +; CHECK-NEXT: ret i32 [[T1_NEG]] ; %t0 = ashr exact i64 %x, 63 %t1 = trunc i64 %t0 to i32 @@ -48,9 +48,9 @@ define <2 x i32> @t4(<2 x i64> %x) { ; CHECK-LABEL: @t4( -; CHECK-NEXT: [[TMP1:%.*]] = ashr <2 x i64> [[X:%.*]], -; CHECK-NEXT: [[R:%.*]] = trunc <2 x i64> [[TMP1]] to <2 x i32> -; CHECK-NEXT: ret <2 x i32> [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr <2 x i64> [[X:%.*]], +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc <2 x i64> [[T0_NEG]] to <2 x i32> +; CHECK-NEXT: ret <2 x i32> [[T1_NEG]] ; %t0 = lshr <2 x i64> %x, %t1 = trunc <2 x i64> %t0 to <2 x i32> @@ -76,11 +76,11 @@ define i32 @t6(i64 %x) { ; CHECK-LABEL: @t6( -; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X]], 63 ; CHECK-NEXT: call void @use64(i64 [[T0]]) -; CHECK-NEXT: [[TMP1:%.*]] = ashr i64 [[X]], 63 -; CHECK-NEXT: [[R:%.*]] = trunc i64 [[TMP1]] to i32 -; CHECK-NEXT: ret i32 [[R]] +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc i64 [[T0_NEG]] to i32 +; CHECK-NEXT: ret i32 [[T1_NEG]] ; %t0 = lshr i64 %x, 63 call void @use64(i64 %t0) @@ -136,9 +136,9 @@ define i32 @n10(i64 %x) { ; CHECK-LABEL: @n10( -; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X:%.*]], 63 -; CHECK-NEXT: [[T1:%.*]] = trunc i64 [[T0]] to i32 -; CHECK-NEXT: [[R:%.*]] = xor i32 [[T1]], 1 +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc i64 [[T0_NEG]] to i32 +; CHECK-NEXT: [[R:%.*]] = add i32 [[T1_NEG]], 1 ; CHECK-NEXT: ret i32 [[R]] ; %t0 = lshr i64 %x, 63 diff --git a/llvm/test/Transforms/InstCombine/high-bit-signmask.ll b/llvm/test/Transforms/InstCombine/high-bit-signmask.ll --- a/llvm/test/Transforms/InstCombine/high-bit-signmask.ll +++ b/llvm/test/Transforms/InstCombine/high-bit-signmask.ll @@ -3,8 +3,8 @@ define i64 @t0(i64 %x) { ; CHECK-LABEL: @t0( -; CHECK-NEXT: [[R:%.*]] = ashr i64 [[X:%.*]], 63 -; CHECK-NEXT: ret i64 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: ret i64 [[T0_NEG]] ; %t0 = lshr i64 %x, 63 %r = sub i64 0, %t0 @@ -12,8 +12,8 @@ } define i64 @t0_exact(i64 %x) { ; CHECK-LABEL: @t0_exact( -; CHECK-NEXT: [[R:%.*]] = ashr exact i64 [[X:%.*]], 63 -; CHECK-NEXT: ret i64 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr exact i64 [[X:%.*]], 63 +; CHECK-NEXT: ret i64 [[T0_NEG]] ; %t0 = lshr exact i64 %x, 63 %r = sub i64 0, %t0 @@ -21,8 +21,8 @@ } define i64 @t2(i64 %x) { ; CHECK-LABEL: @t2( -; CHECK-NEXT: [[R:%.*]] = lshr i64 [[X:%.*]], 63 -; CHECK-NEXT: ret i64 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = lshr i64 [[X:%.*]], 63 +; CHECK-NEXT: ret i64 [[T0_NEG]] ; %t0 = ashr i64 %x, 63 %r = sub i64 0, %t0 @@ -30,8 +30,8 @@ } define i64 @t3_exact(i64 %x) { ; CHECK-LABEL: @t3_exact( -; CHECK-NEXT: [[R:%.*]] = lshr exact i64 [[X:%.*]], 63 -; CHECK-NEXT: ret i64 [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = lshr exact i64 [[X:%.*]], 63 +; CHECK-NEXT: ret i64 [[T0_NEG]] ; %t0 = ashr exact i64 %x, 63 %r = sub i64 0, %t0 @@ -40,8 +40,8 @@ define <2 x i64> @t4(<2 x i64> %x) { ; CHECK-LABEL: @t4( -; CHECK-NEXT: [[R:%.*]] = ashr <2 x i64> [[X:%.*]], -; CHECK-NEXT: ret <2 x i64> [[R]] +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr <2 x i64> [[X:%.*]], +; CHECK-NEXT: ret <2 x i64> [[T0_NEG]] ; %t0 = lshr <2 x i64> %x, %r = sub <2 x i64> zeroinitializer, %t0 @@ -64,10 +64,10 @@ define i64 @t6(i64 %x) { ; CHECK-LABEL: @t6( -; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X]], 63 ; CHECK-NEXT: call void @use64(i64 [[T0]]) -; CHECK-NEXT: [[R:%.*]] = ashr i64 [[X]], 63 -; CHECK-NEXT: ret i64 [[R]] +; CHECK-NEXT: ret i64 [[T0_NEG]] ; %t0 = lshr i64 %x, 63 call void @use64(i64 %t0) @@ -77,10 +77,10 @@ define i64 @n7(i64 %x) { ; CHECK-LABEL: @n7( -; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X]], 63 ; CHECK-NEXT: call void @use32(i64 [[T0]]) -; CHECK-NEXT: [[R:%.*]] = ashr i64 [[X]], 63 -; CHECK-NEXT: ret i64 [[R]] +; CHECK-NEXT: ret i64 [[T0_NEG]] ; %t0 = lshr i64 %x, 63 call void @use32(i64 %t0) @@ -90,11 +90,11 @@ define i64 @n8(i64 %x) { ; CHECK-LABEL: @n8( -; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X]], 63 ; CHECK-NEXT: call void @use64(i64 [[T0]]) ; CHECK-NEXT: call void @use32(i64 [[T0]]) -; CHECK-NEXT: [[R:%.*]] = ashr i64 [[X]], 63 -; CHECK-NEXT: ret i64 [[R]] +; CHECK-NEXT: ret i64 [[T0_NEG]] ; %t0 = lshr i64 %x, 63 call void @use64(i64 %t0) @@ -116,8 +116,8 @@ define i64 @n10(i64 %x) { ; CHECK-LABEL: @n10( -; CHECK-NEXT: [[T0:%.*]] = lshr i64 [[X:%.*]], 63 -; CHECK-NEXT: [[R:%.*]] = xor i64 [[T0]], 1 +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i64 [[X:%.*]], 63 +; CHECK-NEXT: [[R:%.*]] = add nsw i64 [[T0_NEG]], 1 ; CHECK-NEXT: ret i64 [[R]] ; %t0 = lshr i64 %x, 63 diff --git a/llvm/test/Transforms/InstCombine/sub-of-negatible.ll b/llvm/test/Transforms/InstCombine/sub-of-negatible.ll --- a/llvm/test/Transforms/InstCombine/sub-of-negatible.ll +++ b/llvm/test/Transforms/InstCombine/sub-of-negatible.ll @@ -31,8 +31,8 @@ ; Shift-left can be negated if all uses can be updated define i8 @t2(i8 %x, i8 %y) { ; CHECK-LABEL: @t2( -; CHECK-NEXT: [[T0:%.*]] = shl i8 -42, [[Y:%.*]] -; CHECK-NEXT: [[T1:%.*]] = sub i8 [[X:%.*]], [[T0]] +; CHECK-NEXT: [[T0_NEG:%.*]] = shl i8 42, [[Y:%.*]] +; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T1]] ; %t0 = shl i8 -42, %y @@ -55,8 +55,8 @@ ; CHECK-LABEL: @t3( ; CHECK-NEXT: [[T0:%.*]] = sub i8 0, [[Z:%.*]] ; CHECK-NEXT: call void @use8(i8 [[T0]]) -; CHECK-NEXT: [[T1:%.*]] = shl i8 [[T0]], [[Y:%.*]] -; CHECK-NEXT: [[T2:%.*]] = sub i8 [[X:%.*]], [[T1]] +; CHECK-NEXT: [[T1_NEG:%.*]] = shl i8 [[Z]], [[Y:%.*]] +; CHECK-NEXT: [[T2:%.*]] = add i8 [[T1_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T2]] ; %t0 = sub i8 0, %z @@ -85,8 +85,8 @@ ; Select can be negated if all it's operands can be negated and all the users of select can be updated define i8 @t4(i8 %x, i1 %y) { ; CHECK-LABEL: @t4( -; CHECK-NEXT: [[T0:%.*]] = select i1 [[Y:%.*]], i8 -42, i8 44 -; CHECK-NEXT: [[T1:%.*]] = sub i8 [[X:%.*]], [[T0]] +; CHECK-NEXT: [[T0_NEG:%.*]] = select i1 [[Y:%.*]], i8 42, i8 -44 +; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T1]] ; %t0 = select i1 %y, i8 -42, i8 44 @@ -119,8 +119,8 @@ ; CHECK-LABEL: @t6( ; CHECK-NEXT: [[T0:%.*]] = sub i8 0, [[Z:%.*]] ; CHECK-NEXT: call void @use8(i8 [[T0]]) -; CHECK-NEXT: [[T1:%.*]] = select i1 [[Y:%.*]], i8 -42, i8 [[T0]] -; CHECK-NEXT: [[T2:%.*]] = sub i8 [[X:%.*]], [[T1]] +; CHECK-NEXT: [[T1_NEG:%.*]] = select i1 [[Y:%.*]], i8 42, i8 [[Z]] +; CHECK-NEXT: [[T2:%.*]] = add i8 [[T1_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T2]] ; %t0 = sub i8 0, %z @@ -131,9 +131,9 @@ } define i8 @t7(i8 %x, i1 %y, i8 %z) { ; CHECK-LABEL: @t7( -; CHECK-NEXT: [[T0:%.*]] = shl i8 1, [[Z:%.*]] -; CHECK-NEXT: [[T1:%.*]] = select i1 [[Y:%.*]], i8 0, i8 [[T0]] -; CHECK-NEXT: [[T2:%.*]] = sub i8 [[X:%.*]], [[T1]] +; CHECK-NEXT: [[T0_NEG:%.*]] = shl i8 -1, [[Z:%.*]] +; CHECK-NEXT: [[T1_NEG:%.*]] = select i1 [[Y:%.*]], i8 0, i8 [[T0_NEG]] +; CHECK-NEXT: [[T2:%.*]] = add i8 [[T1_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T2]] ; %t0 = shl i8 1, %z @@ -169,10 +169,10 @@ } define i8 @n10(i8 %x, i8 %y, i8 %z) { ; CHECK-LABEL: @n10( -; CHECK-NEXT: [[T0:%.*]] = sub i8 [[Y:%.*]], [[X:%.*]] +; CHECK-NEXT: [[T0_NEG:%.*]] = sub i8 [[X:%.*]], [[Y:%.*]] +; CHECK-NEXT: [[T0:%.*]] = sub i8 [[Y]], [[X]] ; CHECK-NEXT: call void @use8(i8 [[T0]]) -; CHECK-NEXT: [[T1:%.*]] = sub i8 0, [[T0]] -; CHECK-NEXT: ret i8 [[T1]] +; CHECK-NEXT: ret i8 [[T0_NEG]] ; %t0 = sub i8 %y, %x call void @use8(i8 %t0) @@ -204,8 +204,8 @@ ; CHECK-LABEL: @n13( ; CHECK-NEXT: [[T0:%.*]] = sub i8 0, [[Y:%.*]] ; CHECK-NEXT: call void @use8(i8 [[T0]]) -; CHECK-NEXT: [[T11:%.*]] = sub i8 [[Y]], [[Z:%.*]] -; CHECK-NEXT: [[T2:%.*]] = add i8 [[T11]], [[X:%.*]] +; CHECK-NEXT: [[T1_NEG:%.*]] = sub i8 [[Y]], [[Z:%.*]] +; CHECK-NEXT: [[T2:%.*]] = add i8 [[T1_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T2]] ; %t0 = sub i8 0, %y @@ -242,8 +242,8 @@ ; CHECK-LABEL: @t15( ; CHECK-NEXT: [[T0:%.*]] = sub i8 0, [[Y:%.*]] ; CHECK-NEXT: call void @use8(i8 [[T0]]) -; CHECK-NEXT: [[TMP1:%.*]] = mul i8 [[Z:%.*]], [[Y]] -; CHECK-NEXT: [[T2:%.*]] = add i8 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[T1_NEG:%.*]] = mul i8 [[Y]], [[Z:%.*]] +; CHECK-NEXT: [[T2:%.*]] = add i8 [[T1_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T2]] ; %t0 = sub i8 0, %y @@ -279,8 +279,8 @@ ; CHECK: else: ; CHECK-NEXT: br label [[END]] ; CHECK: end: -; CHECK-NEXT: [[Z:%.*]] = phi i8 [ [[X:%.*]], [[THEN]] ], [ 42, [[ELSE]] ] -; CHECK-NEXT: ret i8 [[Z]] +; CHECK-NEXT: [[Z_NEG:%.*]] = phi i8 [ [[X:%.*]], [[THEN]] ], [ 42, [[ELSE]] ] +; CHECK-NEXT: ret i8 [[Z_NEG]] ; begin: br i1 %c, label %then, label %else @@ -352,9 +352,9 @@ ; truncation can be negated if it's operand can be negated define i8 @t20(i8 %x, i16 %y) { ; CHECK-LABEL: @t20( -; CHECK-NEXT: [[T0:%.*]] = shl i16 -42, [[Y:%.*]] -; CHECK-NEXT: [[T1:%.*]] = trunc i16 [[T0]] to i8 -; CHECK-NEXT: [[T2:%.*]] = sub i8 [[X:%.*]], [[T1]] +; CHECK-NEXT: [[T0_NEG:%.*]] = shl i16 42, [[Y:%.*]] +; CHECK-NEXT: [[T1_NEG:%.*]] = trunc i16 [[T0_NEG]] to i8 +; CHECK-NEXT: [[T2:%.*]] = add i8 [[T1_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T2]] ; %t0 = shl i16 -42, %y @@ -427,9 +427,9 @@ define i8 @negate_shl_not_uses(i8 %x, i8 %y) { ; CHECK-LABEL: @negate_shl_not_uses( -; CHECK-NEXT: [[O:%.*]] = xor i8 [[X:%.*]], -1 +; CHECK-NEXT: [[O_NEG:%.*]] = add i8 [[X:%.*]], 1 +; CHECK-NEXT: [[O:%.*]] = xor i8 [[X]], -1 ; CHECK-NEXT: call void @use8(i8 [[O]]) -; CHECK-NEXT: [[O_NEG:%.*]] = add i8 [[X]], 1 ; CHECK-NEXT: [[S_NEG:%.*]] = shl i8 [[O_NEG]], [[Y:%.*]] ; CHECK-NEXT: ret i8 [[S_NEG]] ; @@ -442,9 +442,9 @@ define <2 x i4> @negate_mul_not_uses_vec(<2 x i4> %x, <2 x i4> %y) { ; CHECK-LABEL: @negate_mul_not_uses_vec( -; CHECK-NEXT: [[O:%.*]] = xor <2 x i4> [[X:%.*]], +; CHECK-NEXT: [[O_NEG:%.*]] = add <2 x i4> [[X:%.*]], +; CHECK-NEXT: [[O:%.*]] = xor <2 x i4> [[X]], ; CHECK-NEXT: call void @use_v2i4(<2 x i4> [[O]]) -; CHECK-NEXT: [[O_NEG:%.*]] = add <2 x i4> [[X]], ; CHECK-NEXT: [[S_NEG:%.*]] = mul <2 x i4> [[O_NEG]], [[Y:%.*]] ; CHECK-NEXT: ret <2 x i4> [[S_NEG]] ; @@ -458,8 +458,8 @@ ; signed division can be negated if divisor can be negated and is not 1/-1 define i8 @negate_sdiv(i8 %x, i8 %y) { ; CHECK-LABEL: @negate_sdiv( -; CHECK-NEXT: [[T0:%.*]] = sdiv i8 [[Y:%.*]], 42 -; CHECK-NEXT: [[T1:%.*]] = sub i8 [[X:%.*]], [[T0]] +; CHECK-NEXT: [[T0_NEG:%.*]] = sdiv i8 [[Y:%.*]], -42 +; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T1]] ; %t0 = sdiv i8 %y, 42 @@ -478,12 +478,24 @@ %t1 = sub i8 %x, %t0 ret i8 %t1 } +define i8 @negate_sdiv_extrause2(i8 %x, i8 %y) { +; CHECK-LABEL: @negate_sdiv_extrause2( +; CHECK-NEXT: [[T0:%.*]] = sdiv i8 [[Y:%.*]], 42 +; CHECK-NEXT: call void @use8(i8 [[T0]]) +; CHECK-NEXT: [[T1:%.*]] = sub nsw i8 0, [[T0]] +; CHECK-NEXT: ret i8 [[T1]] +; + %t0 = sdiv i8 %y, 42 + call void @use8(i8 %t0) + %t1 = sub i8 0, %t0 + ret i8 %t1 +} ; Right-shift sign bit smear is negatible. define i8 @negate_ashr(i8 %x, i8 %y) { ; CHECK-LABEL: @negate_ashr( -; CHECK-NEXT: [[T0:%.*]] = ashr i8 [[Y:%.*]], 7 -; CHECK-NEXT: [[T1:%.*]] = sub i8 [[X:%.*]], [[T0]] +; CHECK-NEXT: [[T0_NEG:%.*]] = lshr i8 [[Y:%.*]], 7 +; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T1]] ; %t0 = ashr i8 %y, 7 @@ -492,8 +504,8 @@ } define i8 @negate_lshr(i8 %x, i8 %y) { ; CHECK-LABEL: @negate_lshr( -; CHECK-NEXT: [[T0:%.*]] = lshr i8 [[Y:%.*]], 7 -; CHECK-NEXT: [[T1:%.*]] = sub i8 [[X:%.*]], [[T0]] +; CHECK-NEXT: [[T0_NEG:%.*]] = ashr i8 [[Y:%.*]], 7 +; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T1]] ; %t0 = lshr i8 %y, 7 @@ -548,8 +560,8 @@ ; *ext of i1 is always negatible define i8 @negate_sext(i8 %x, i1 %y) { ; CHECK-LABEL: @negate_sext( -; CHECK-NEXT: [[TMP1:%.*]] = zext i1 [[Y:%.*]] to i8 -; CHECK-NEXT: [[T1:%.*]] = add i8 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[T0_NEG:%.*]] = zext i1 [[Y:%.*]] to i8 +; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T1]] ; %t0 = sext i1 %y to i8 @@ -558,8 +570,8 @@ } define i8 @negate_zext(i8 %x, i1 %y) { ; CHECK-LABEL: @negate_zext( -; CHECK-NEXT: [[TMP1:%.*]] = sext i1 [[Y:%.*]] to i8 -; CHECK-NEXT: [[T1:%.*]] = add i8 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[T0_NEG:%.*]] = sext i1 [[Y:%.*]] to i8 +; CHECK-NEXT: [[T1:%.*]] = add i8 [[T0_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[T1]] ; %t0 = zext i1 %y to i8 diff --git a/llvm/test/Transforms/InstCombine/sub.ll b/llvm/test/Transforms/InstCombine/sub.ll --- a/llvm/test/Transforms/InstCombine/sub.ll +++ b/llvm/test/Transforms/InstCombine/sub.ll @@ -218,8 +218,8 @@ define i32 @test5(i32 %A, i32 %B, i32 %C) { ; CHECK-LABEL: @test5( -; CHECK-NEXT: [[D1:%.*]] = sub i32 [[C:%.*]], [[B:%.*]] -; CHECK-NEXT: [[E:%.*]] = add i32 [[D1]], [[A:%.*]] +; CHECK-NEXT: [[D_NEG:%.*]] = sub i32 [[C:%.*]], [[B:%.*]] +; CHECK-NEXT: [[E:%.*]] = add i32 [[D_NEG]], [[A:%.*]] ; CHECK-NEXT: ret i32 [[E]] ; %D = sub i32 %B, %C @@ -555,11 +555,11 @@ define i64 @test_neg_shl_sub_extra_use1(i64 %a, i64 %b, i64* %p) { ; CHECK-LABEL: @test_neg_shl_sub_extra_use1( -; CHECK-NEXT: [[SUB:%.*]] = sub i64 [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[SUB_NEG:%.*]] = sub i64 [[B:%.*]], [[A:%.*]] +; CHECK-NEXT: [[SUB:%.*]] = sub i64 [[A]], [[B]] ; CHECK-NEXT: store i64 [[SUB]], i64* [[P:%.*]], align 8 -; CHECK-NEXT: [[MUL:%.*]] = shl i64 [[SUB]], 2 -; CHECK-NEXT: [[NEG:%.*]] = sub i64 0, [[MUL]] -; CHECK-NEXT: ret i64 [[NEG]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = shl i64 [[SUB_NEG]], 2 +; CHECK-NEXT: ret i64 [[MUL_NEG]] ; %sub = sub i64 %a, %b store i64 %sub, i64* %p @@ -621,8 +621,8 @@ define i64 @test_neg_zext_i1_extra_use(i1 %a, i64 %b, i64* %p) { ; CHECK-LABEL: @test_neg_zext_i1_extra_use( -; CHECK-NEXT: [[EXT:%.*]] = zext i1 [[A:%.*]] to i64 -; CHECK-NEXT: [[EXT_NEG:%.*]] = sext i1 [[A]] to i64 +; CHECK-NEXT: [[EXT_NEG:%.*]] = sext i1 [[A:%.*]] to i64 +; CHECK-NEXT: [[EXT:%.*]] = zext i1 [[A]] to i64 ; CHECK-NEXT: store i64 [[EXT]], i64* [[P:%.*]], align 8 ; CHECK-NEXT: ret i64 [[EXT_NEG]] ; @@ -634,8 +634,8 @@ define i64 @test_neg_sext_i1_extra_use(i1 %a, i64 %b, i64* %p) { ; CHECK-LABEL: @test_neg_sext_i1_extra_use( -; CHECK-NEXT: [[EXT:%.*]] = sext i1 [[A:%.*]] to i64 -; CHECK-NEXT: [[EXT_NEG:%.*]] = zext i1 [[A]] to i64 +; CHECK-NEXT: [[EXT_NEG:%.*]] = zext i1 [[A:%.*]] to i64 +; CHECK-NEXT: [[EXT:%.*]] = sext i1 [[A]] to i64 ; CHECK-NEXT: store i64 [[EXT]], i64* [[P:%.*]], align 8 ; CHECK-NEXT: ret i64 [[EXT_NEG]] ; @@ -703,7 +703,7 @@ ; CHECK-LABEL: @test_neg_mul_sub_commuted( ; CHECK-NEXT: [[COMPLEX:%.*]] = mul i64 [[C:%.*]], [[C]] ; CHECK-NEXT: [[SUB_NEG:%.*]] = sub i64 [[B:%.*]], [[A:%.*]] -; CHECK-NEXT: [[MUL_NEG:%.*]] = mul i64 [[COMPLEX]], [[SUB_NEG]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = mul i64 [[SUB_NEG]], [[COMPLEX]] ; CHECK-NEXT: ret i64 [[MUL_NEG]] ; %complex = mul i64 %c, %c @@ -715,8 +715,8 @@ define i32 @test27(i32 %x, i32 %y) { ; CHECK-LABEL: @test27( -; CHECK-NEXT: [[TMP1:%.*]] = shl i32 [[Y:%.*]], 3 -; CHECK-NEXT: [[SUB:%.*]] = add i32 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = shl i32 [[Y:%.*]], 3 +; CHECK-NEXT: [[SUB:%.*]] = add i32 [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i32 [[SUB]] ; %mul = mul i32 %y, -8 @@ -726,8 +726,8 @@ define <2 x i32> @test27vec(<2 x i32> %x, <2 x i32> %y) { ; CHECK-LABEL: @test27vec( -; CHECK-NEXT: [[TMP1:%.*]] = mul <2 x i32> [[Y:%.*]], -; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = mul <2 x i32> [[Y:%.*]], +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i32> [[SUB]] ; %mul = mul <2 x i32> %y, @@ -737,8 +737,8 @@ define <2 x i32> @test27vecsplat(<2 x i32> %x, <2 x i32> %y) { ; CHECK-LABEL: @test27vecsplat( -; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> [[Y:%.*]], -; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = shl <2 x i32> [[Y:%.*]], +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i32> [[SUB]] ; %mul = mul <2 x i32> %y, @@ -748,8 +748,8 @@ define <2 x i32> @test27vecmixed(<2 x i32> %x, <2 x i32> %y) { ; CHECK-LABEL: @test27vecmixed( -; CHECK-NEXT: [[TMP1:%.*]] = mul <2 x i32> [[Y:%.*]], -; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = mul <2 x i32> [[Y:%.*]], +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i32> [[SUB]] ; %mul = mul <2 x i32> %y, @@ -759,8 +759,8 @@ define i32 @test27commuted(i32 %x, i32 %y) { ; CHECK-LABEL: @test27commuted( -; CHECK-NEXT: [[TMP1:%.*]] = shl i32 [[Y:%.*]], 3 -; CHECK-NEXT: [[SUB:%.*]] = add i32 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = shl i32 [[Y:%.*]], 3 +; CHECK-NEXT: [[SUB:%.*]] = add i32 [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i32 [[SUB]] ; %mul = mul i32 -8, %y @@ -770,8 +770,8 @@ define <2 x i32> @test27commutedvec(<2 x i32> %x, <2 x i32> %y) { ; CHECK-LABEL: @test27commutedvec( -; CHECK-NEXT: [[TMP1:%.*]] = mul <2 x i32> [[Y:%.*]], -; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = mul <2 x i32> [[Y:%.*]], +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i32> [[SUB]] ; %mul = mul <2 x i32> , %y @@ -781,8 +781,8 @@ define <2 x i32> @test27commutedvecsplat(<2 x i32> %x, <2 x i32> %y) { ; CHECK-LABEL: @test27commutedvecsplat( -; CHECK-NEXT: [[TMP1:%.*]] = shl <2 x i32> [[Y:%.*]], -; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = shl <2 x i32> [[Y:%.*]], +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i32> [[SUB]] ; %mul = mul <2 x i32> , %y @@ -792,8 +792,8 @@ define <2 x i32> @test27commutedvecmixed(<2 x i32> %x, <2 x i32> %y) { ; CHECK-LABEL: @test27commutedvecmixed( -; CHECK-NEXT: [[TMP1:%.*]] = mul <2 x i32> [[Y:%.*]], -; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[MUL_NEG:%.*]] = mul <2 x i32> [[Y:%.*]], +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i32> [[MUL_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i32> [[SUB]] ; %mul = mul <2 x i32> , %y @@ -1105,7 +1105,7 @@ define <2 x i1> @test53(<2 x i1> %A, <2 x i1> %B) { ; CHECK-LABEL: @test53( -; CHECK-NEXT: [[SUB:%.*]] = xor <2 x i1> [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[SUB:%.*]] = xor <2 x i1> [[B:%.*]], [[A:%.*]] ; CHECK-NEXT: ret <2 x i1> [[SUB]] ; %sub = sub <2 x i1> %A, %B @@ -1149,8 +1149,8 @@ ; CHECK: delay: ; CHECK-NEXT: br label [[FINAL]] ; CHECK: final: -; CHECK-NEXT: [[A:%.*]] = phi i32 [ -877, [[ENTRY:%.*]] ], [ 113, [[DELAY]] ] -; CHECK-NEXT: ret i32 [[A]] +; CHECK-NEXT: [[A_NEG:%.*]] = phi i32 [ -877, [[ENTRY:%.*]] ], [ 113, [[DELAY]] ] +; CHECK-NEXT: ret i32 [[A_NEG]] ; entry: br i1 %which, label %final, label %delay @@ -1171,8 +1171,8 @@ ; CHECK: delay: ; CHECK-NEXT: br label [[FINAL]] ; CHECK: final: -; CHECK-NEXT: [[A:%.*]] = phi <2 x i32> [ , [[ENTRY:%.*]] ], [ , [[DELAY]] ] -; CHECK-NEXT: ret <2 x i32> [[A]] +; CHECK-NEXT: [[A_NEG:%.*]] = phi <2 x i32> [ , [[ENTRY:%.*]] ], [ , [[DELAY]] ] +; CHECK-NEXT: ret <2 x i32> [[A_NEG]] ; entry: br i1 %which, label %final, label %delay @@ -1193,8 +1193,8 @@ ; CHECK: delay: ; CHECK-NEXT: br label [[FINAL]] ; CHECK: final: -; CHECK-NEXT: [[A:%.*]] = phi <2 x i32> [ , [[ENTRY:%.*]] ], [ , [[DELAY]] ] -; CHECK-NEXT: ret <2 x i32> [[A]] +; CHECK-NEXT: [[A_NEG:%.*]] = phi <2 x i32> [ , [[ENTRY:%.*]] ], [ , [[DELAY]] ] +; CHECK-NEXT: ret <2 x i32> [[A_NEG]] ; entry: br i1 %which, label %final, label %delay @@ -1356,8 +1356,8 @@ ; CHECK-LABEL: @test64( ; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[X:%.*]], 255 ; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[X]], i32 255 -; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[TMP2]], 1 -; CHECK-NEXT: ret i32 [[RES]] +; CHECK-NEXT: [[DOTNEG:%.*]] = add nsw i32 [[TMP2]], 1 +; CHECK-NEXT: ret i32 [[DOTNEG]] ; %1 = xor i32 %x, -1 %2 = icmp sgt i32 %1, -256 @@ -1370,8 +1370,8 @@ ; CHECK-LABEL: @test65( ; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[X:%.*]], -256 ; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[X]], i32 -256 -; CHECK-NEXT: [[RES:%.*]] = add i32 [[TMP2]], 1 -; CHECK-NEXT: ret i32 [[RES]] +; CHECK-NEXT: [[DOTNEG:%.*]] = add i32 [[TMP2]], 1 +; CHECK-NEXT: ret i32 [[DOTNEG]] ; %1 = xor i32 %x, -1 %2 = icmp slt i32 %1, 255 @@ -1384,8 +1384,8 @@ ; CHECK-LABEL: @test66( ; CHECK-NEXT: [[TMP1:%.*]] = icmp ult i32 [[X:%.*]], -101 ; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[X]], i32 -101 -; CHECK-NEXT: [[RES:%.*]] = add nuw i32 [[TMP2]], 1 -; CHECK-NEXT: ret i32 [[RES]] +; CHECK-NEXT: [[DOTNEG:%.*]] = add nuw i32 [[TMP2]], 1 +; CHECK-NEXT: ret i32 [[DOTNEG]] ; %1 = xor i32 %x, -1 %2 = icmp ugt i32 %1, 100 @@ -1398,8 +1398,8 @@ ; CHECK-LABEL: @test67( ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i32 [[X:%.*]], 100 ; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[X]], i32 100 -; CHECK-NEXT: [[RES:%.*]] = add i32 [[TMP2]], 1 -; CHECK-NEXT: ret i32 [[RES]] +; CHECK-NEXT: [[DOTNEG:%.*]] = add i32 [[TMP2]], 1 +; CHECK-NEXT: ret i32 [[DOTNEG]] ; %1 = xor i32 %x, -1 %2 = icmp ult i32 %1, -101 @@ -1413,8 +1413,8 @@ ; CHECK-LABEL: @test68( ; CHECK-NEXT: [[TMP1:%.*]] = icmp slt <2 x i32> [[X:%.*]], ; CHECK-NEXT: [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x i32> [[X]], <2 x i32> -; CHECK-NEXT: [[RES:%.*]] = add nsw <2 x i32> [[TMP2]], -; CHECK-NEXT: ret <2 x i32> [[RES]] +; CHECK-NEXT: [[DOTNEG:%.*]] = add nsw <2 x i32> [[TMP2]], +; CHECK-NEXT: ret <2 x i32> [[DOTNEG]] ; %1 = xor <2 x i32> %x, %2 = icmp sgt <2 x i32> %1, @@ -1428,8 +1428,8 @@ ; CHECK-LABEL: @test69( ; CHECK-NEXT: [[TMP1:%.*]] = icmp slt <2 x i32> [[X:%.*]], ; CHECK-NEXT: [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x i32> [[X]], <2 x i32> -; CHECK-NEXT: [[RES:%.*]] = add <2 x i32> [[TMP2]], -; CHECK-NEXT: ret <2 x i32> [[RES]] +; CHECK-NEXT: [[DOTNEG:%.*]] = add <2 x i32> [[TMP2]], +; CHECK-NEXT: ret <2 x i32> [[DOTNEG]] ; %1 = xor <2 x i32> %x, %2 = icmp sgt <2 x i32> %1, diff --git a/llvm/test/Transforms/InstCombine/zext-bool-add-sub.ll b/llvm/test/Transforms/InstCombine/zext-bool-add-sub.ll --- a/llvm/test/Transforms/InstCombine/zext-bool-add-sub.ll +++ b/llvm/test/Transforms/InstCombine/zext-bool-add-sub.ll @@ -5,9 +5,9 @@ define i32 @a(i1 zeroext %x, i1 zeroext %y) { ; CHECK-LABEL: @a( -; CHECK-NEXT: [[CONV3_NEG:%.*]] = sext i1 [[Y:%.*]] to i32 +; CHECK-NEXT: [[CONV3_NEG1:%.*]] = sext i1 [[Y:%.*]] to i32 ; CHECK-NEXT: [[SUB:%.*]] = select i1 [[X:%.*]], i32 2, i32 1 -; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[SUB]], [[CONV3_NEG]] +; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[SUB]], [[CONV3_NEG1]] ; CHECK-NEXT: ret i32 [[ADD]] ; %conv = zext i1 %x to i32 @@ -95,8 +95,8 @@ define i64 @zext_negate(i1 %A) { ; CHECK-LABEL: @zext_negate( -; CHECK-NEXT: [[SUB:%.*]] = sext i1 [[A:%.*]] to i64 -; CHECK-NEXT: ret i64 [[SUB]] +; CHECK-NEXT: [[EXT_NEG:%.*]] = sext i1 [[A:%.*]] to i64 +; CHECK-NEXT: ret i64 [[EXT_NEG]] ; %ext = zext i1 %A to i64 %sub = sub i64 0, %ext @@ -105,10 +105,10 @@ define i64 @zext_negate_extra_use(i1 %A) { ; CHECK-LABEL: @zext_negate_extra_use( -; CHECK-NEXT: [[EXT:%.*]] = zext i1 [[A:%.*]] to i64 -; CHECK-NEXT: [[SUB:%.*]] = sext i1 [[A]] to i64 +; CHECK-NEXT: [[EXT_NEG:%.*]] = sext i1 [[A:%.*]] to i64 +; CHECK-NEXT: [[EXT:%.*]] = zext i1 [[A]] to i64 ; CHECK-NEXT: call void @use(i64 [[EXT]]) -; CHECK-NEXT: ret i64 [[SUB]] +; CHECK-NEXT: ret i64 [[EXT_NEG]] ; %ext = zext i1 %A to i64 %sub = sub i64 0, %ext @@ -118,8 +118,8 @@ define <2 x i64> @zext_negate_vec(<2 x i1> %A) { ; CHECK-LABEL: @zext_negate_vec( -; CHECK-NEXT: [[SUB:%.*]] = sext <2 x i1> [[A:%.*]] to <2 x i64> -; CHECK-NEXT: ret <2 x i64> [[SUB]] +; CHECK-NEXT: [[EXT_NEG:%.*]] = sext <2 x i1> [[A:%.*]] to <2 x i64> +; CHECK-NEXT: ret <2 x i64> [[EXT_NEG]] ; %ext = zext <2 x i1> %A to <2 x i64> %sub = sub <2 x i64> zeroinitializer, %ext @@ -128,8 +128,8 @@ define <2 x i64> @zext_negate_vec_undef_elt(<2 x i1> %A) { ; CHECK-LABEL: @zext_negate_vec_undef_elt( -; CHECK-NEXT: [[SUB:%.*]] = sext <2 x i1> [[A:%.*]] to <2 x i64> -; CHECK-NEXT: ret <2 x i64> [[SUB]] +; CHECK-NEXT: [[EXT_NEG:%.*]] = sext <2 x i1> [[A:%.*]] to <2 x i64> +; CHECK-NEXT: ret <2 x i64> [[EXT_NEG]] ; %ext = zext <2 x i1> %A to <2 x i64> %sub = sub <2 x i64> , %ext @@ -181,8 +181,8 @@ define i64 @sext_negate(i1 %A) { ; CHECK-LABEL: @sext_negate( -; CHECK-NEXT: [[SUB:%.*]] = zext i1 [[A:%.*]] to i64 -; CHECK-NEXT: ret i64 [[SUB]] +; CHECK-NEXT: [[EXT_NEG:%.*]] = zext i1 [[A:%.*]] to i64 +; CHECK-NEXT: ret i64 [[EXT_NEG]] ; %ext = sext i1 %A to i64 %sub = sub i64 0, %ext @@ -191,10 +191,10 @@ define i64 @sext_negate_extra_use(i1 %A) { ; CHECK-LABEL: @sext_negate_extra_use( -; CHECK-NEXT: [[EXT:%.*]] = sext i1 [[A:%.*]] to i64 -; CHECK-NEXT: [[SUB:%.*]] = zext i1 [[A]] to i64 +; CHECK-NEXT: [[EXT_NEG:%.*]] = zext i1 [[A:%.*]] to i64 +; CHECK-NEXT: [[EXT:%.*]] = sext i1 [[A]] to i64 ; CHECK-NEXT: call void @use(i64 [[EXT]]) -; CHECK-NEXT: ret i64 [[SUB]] +; CHECK-NEXT: ret i64 [[EXT_NEG]] ; %ext = sext i1 %A to i64 %sub = sub i64 0, %ext @@ -204,8 +204,8 @@ define <2 x i64> @sext_negate_vec(<2 x i1> %A) { ; CHECK-LABEL: @sext_negate_vec( -; CHECK-NEXT: [[SUB:%.*]] = zext <2 x i1> [[A:%.*]] to <2 x i64> -; CHECK-NEXT: ret <2 x i64> [[SUB]] +; CHECK-NEXT: [[EXT_NEG:%.*]] = zext <2 x i1> [[A:%.*]] to <2 x i64> +; CHECK-NEXT: ret <2 x i64> [[EXT_NEG]] ; %ext = sext <2 x i1> %A to <2 x i64> %sub = sub <2 x i64> zeroinitializer, %ext @@ -214,8 +214,8 @@ define <2 x i64> @sext_negate_vec_undef_elt(<2 x i1> %A) { ; CHECK-LABEL: @sext_negate_vec_undef_elt( -; CHECK-NEXT: [[SUB:%.*]] = zext <2 x i1> [[A:%.*]] to <2 x i64> -; CHECK-NEXT: ret <2 x i64> [[SUB]] +; CHECK-NEXT: [[EXT_NEG:%.*]] = zext <2 x i1> [[A:%.*]] to <2 x i64> +; CHECK-NEXT: ret <2 x i64> [[EXT_NEG]] ; %ext = sext <2 x i1> %A to <2 x i64> %sub = sub <2 x i64> , %ext @@ -267,8 +267,8 @@ define i8 @sext_sub(i8 %x, i1 %y) { ; CHECK-LABEL: @sext_sub( -; CHECK-NEXT: [[TMP1:%.*]] = zext i1 [[Y:%.*]] to i8 -; CHECK-NEXT: [[SUB:%.*]] = add i8 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[SEXT_NEG:%.*]] = zext i1 [[Y:%.*]] to i8 +; CHECK-NEXT: [[SUB:%.*]] = add i8 [[SEXT_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[SUB]] ; %sext = sext i1 %y to i8 @@ -280,8 +280,8 @@ define <2 x i8> @sext_sub_vec(<2 x i8> %x, <2 x i1> %y) { ; CHECK-LABEL: @sext_sub_vec( -; CHECK-NEXT: [[TMP1:%.*]] = zext <2 x i1> [[Y:%.*]] to <2 x i8> -; CHECK-NEXT: [[SUB:%.*]] = add <2 x i8> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[SEXT_NEG:%.*]] = zext <2 x i1> [[Y:%.*]] to <2 x i8> +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i8> [[SEXT_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i8> [[SUB]] ; %sext = sext <2 x i1> %y to <2 x i8> @@ -293,8 +293,8 @@ define <2 x i8> @sext_sub_vec_nsw(<2 x i8> %x, <2 x i1> %y) { ; CHECK-LABEL: @sext_sub_vec_nsw( -; CHECK-NEXT: [[TMP1:%.*]] = zext <2 x i1> [[Y:%.*]] to <2 x i8> -; CHECK-NEXT: [[SUB:%.*]] = add nsw <2 x i8> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[SEXT_NEG:%.*]] = zext <2 x i1> [[Y:%.*]] to <2 x i8> +; CHECK-NEXT: [[SUB:%.*]] = add <2 x i8> [[SEXT_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <2 x i8> [[SUB]] ; %sext = sext <2 x i1> %y to <2 x i8> @@ -306,8 +306,8 @@ define i8 @sext_sub_nuw(i8 %x, i1 %y) { ; CHECK-LABEL: @sext_sub_nuw( -; CHECK-NEXT: [[TMP1:%.*]] = zext i1 [[Y:%.*]] to i8 -; CHECK-NEXT: [[SUB:%.*]] = add i8 [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[SEXT_NEG:%.*]] = zext i1 [[Y:%.*]] to i8 +; CHECK-NEXT: [[SUB:%.*]] = add i8 [[SEXT_NEG]], [[X:%.*]] ; CHECK-NEXT: ret i8 [[SUB]] ; %sext = sext i1 %y to i8 @@ -393,8 +393,8 @@ define <4 x i32> @zextbool_sub_vector(<4 x i1> %c, <4 x i32> %x) { ; CHECK-LABEL: @zextbool_sub_vector( -; CHECK-NEXT: [[TMP1:%.*]] = sext <4 x i1> [[C:%.*]] to <4 x i32> -; CHECK-NEXT: [[S:%.*]] = add <4 x i32> [[TMP1]], [[X:%.*]] +; CHECK-NEXT: [[B_NEG:%.*]] = sext <4 x i1> [[C:%.*]] to <4 x i32> +; CHECK-NEXT: [[S:%.*]] = add <4 x i32> [[B_NEG]], [[X:%.*]] ; CHECK-NEXT: ret <4 x i32> [[S]] ; %b = zext <4 x i1> %c to <4 x i32>