diff options
Diffstat (limited to 'lib/Transforms')
| -rw-r--r-- | lib/Transforms/InstCombine/InstCombine.h | 50 | ||||
| -rw-r--r-- | lib/Transforms/InstCombine/InstCombineAddSub.cpp | 4 | ||||
| -rw-r--r-- | lib/Transforms/InstCombine/InstCombineCasts.cpp | 2 | ||||
| -rw-r--r-- | lib/Transforms/InstCombine/InstCombineMulDivRem.cpp | 226 | ||||
| -rw-r--r-- | lib/Transforms/InstCombine/InstCombineWorklist.h | 30 | ||||
| -rw-r--r-- | lib/Transforms/Instrumentation/MemorySanitizer.cpp | 15 | ||||
| -rw-r--r-- | lib/Transforms/Utils/BasicBlockUtils.cpp | 76 | ||||
| -rw-r--r-- | lib/Transforms/Vectorize/LoopVectorize.cpp | 2 |
8 files changed, 279 insertions, 126 deletions
diff --git a/lib/Transforms/InstCombine/InstCombine.h b/lib/Transforms/InstCombine/InstCombine.h index a36b1e6b9e..1f6a3a5e33 100644 --- a/lib/Transforms/InstCombine/InstCombine.h +++ b/lib/Transforms/InstCombine/InstCombine.h @@ -27,7 +27,7 @@ namespace llvm { class DbgDeclareInst; class MemIntrinsic; class MemSetInst; - + /// SelectPatternFlavor - We can match a variety of different patterns for /// select operations. enum SelectPatternFlavor { @@ -36,7 +36,7 @@ enum SelectPatternFlavor { SPF_SMAX, SPF_UMAX //SPF_ABS - TODO. }; - + /// getComplexity: Assign a complexity or rank value to LLVM Values... /// 0 -> undef, 1 -> Const, 2 -> Other, 3 -> Arg, 3 -> Unary, 4 -> OtherInst static inline unsigned getComplexity(Value *V) { @@ -51,23 +51,23 @@ static inline unsigned getComplexity(Value *V) { return isa<Constant>(V) ? (isa<UndefValue>(V) ? 0 : 1) : 2; } - + /// InstCombineIRInserter - This is an IRBuilder insertion helper that works /// just like the normal insertion helper, but also adds any new instructions /// to the instcombine worklist. -class LLVM_LIBRARY_VISIBILITY InstCombineIRInserter +class LLVM_LIBRARY_VISIBILITY InstCombineIRInserter : public IRBuilderDefaultInserter<true> { InstCombineWorklist &Worklist; public: InstCombineIRInserter(InstCombineWorklist &WL) : Worklist(WL) {} - + void InsertHelper(Instruction *I, const Twine &Name, BasicBlock *BB, BasicBlock::iterator InsertPt) const { IRBuilderDefaultInserter<true>::InsertHelper(I, Name, BB, InsertPt); Worklist.Add(I); } }; - + /// InstCombiner - The -instcombine pass. class LLVM_LIBRARY_VISIBILITY InstCombiner : public FunctionPass, @@ -85,7 +85,7 @@ public: /// instructions into the worklist when they are created. typedef IRBuilder<true, TargetFolder, InstCombineIRInserter> BuilderTy; BuilderTy *Builder; - + static char ID; // Pass identification, replacement for typeid InstCombiner() : FunctionPass(ID), TD(0), Builder(0) { MinimizeSize = false; @@ -94,7 +94,7 @@ public: public: virtual bool runOnFunction(Function &F); - + bool DoOneIteration(Function &F, unsigned ItNum); virtual void getAnalysisUsage(AnalysisUsage &AU) const; @@ -212,10 +212,10 @@ private: bool ShouldChangeType(Type *From, Type *To) const; Value *dyn_castNegVal(Value *V) const; Value *dyn_castFNegVal(Value *V, bool NoSignedZero=false) const; - Type *FindElementAtOffset(Type *Ty, int64_t Offset, + Type *FindElementAtOffset(Type *Ty, int64_t Offset, SmallVectorImpl<Value*> &NewIndices); Instruction *FoldOpIntoSelect(Instruction &Op, SelectInst *SI); - + /// ShouldOptimizeCast - Return true if the cast from "V to Ty" actually /// results in any code being generated and is interesting to optimize out. If /// the cast can be eliminated by some other simple transformation, we prefer @@ -247,7 +247,7 @@ public: return New; } - // InsertNewInstWith - same as InsertNewInstBefore, but also sets the + // InsertNewInstWith - same as InsertNewInstBefore, but also sets the // debug loc. // Instruction *InsertNewInstWith(Instruction *New, Instruction &Old) { @@ -263,10 +263,10 @@ public: // Instruction *ReplaceInstUsesWith(Instruction &I, Value *V) { Worklist.AddUsersToWorkList(I); // Add all modified instrs to worklist. - + // If we are replacing the instruction with itself, this must be in a // segment of unreachable code, so just clobber the instruction. - if (&I == V) + if (&I == V) V = UndefValue::get(I.getType()); DEBUG(errs() << "IC: Replacing " << I << "\n" @@ -296,13 +296,13 @@ public: MadeIRChange = true; return 0; // Don't do anything with FI } - + void ComputeMaskedBits(Value *V, APInt &KnownZero, APInt &KnownOne, unsigned Depth = 0) const { return llvm::ComputeMaskedBits(V, KnownZero, KnownOne, TD, Depth); } - - bool MaskedValueIsZero(Value *V, const APInt &Mask, + + bool MaskedValueIsZero(Value *V, const APInt &Mask, unsigned Depth = 0) const { return llvm::MaskedValueIsZero(V, Mask, TD, Depth); } @@ -325,10 +325,10 @@ private: /// SimplifyDemandedUseBits - Attempts to replace V with a simpler value /// based on the demanded bits. - Value *SimplifyDemandedUseBits(Value *V, APInt DemandedMask, + Value *SimplifyDemandedUseBits(Value *V, APInt DemandedMask, APInt& KnownZero, APInt& KnownOne, unsigned Depth); - bool SimplifyDemandedBits(Use &U, APInt DemandedMask, + bool SimplifyDemandedBits(Use &U, APInt DemandedMask, APInt& KnownZero, APInt& KnownOne, unsigned Depth=0); /// Helper routine of SimplifyDemandedUseBits. It tries to simplify demanded @@ -336,15 +336,15 @@ private: Value *SimplifyShrShlDemandedBits(Instruction *Lsr, Instruction *Sftl, APInt DemandedMask, APInt &KnownZero, APInt &KnownOne); - + /// SimplifyDemandedInstructionBits - Inst is an integer instruction that /// SimplifyDemandedBits knows about. See if the instruction has any /// properties that allow us to simplify its operands. bool SimplifyDemandedInstructionBits(Instruction &Inst); - + Value *SimplifyDemandedVectorElts(Value *V, APInt DemandedElts, APInt& UndefElts, unsigned Depth = 0); - + // FoldOpIntoPhi - Given a binary operator, cast instruction, or select // which has a PHI node as operand #0, see if we can fold the instruction // into the PHI (which is only possible if all operands to the PHI are @@ -360,10 +360,10 @@ private: Instruction *FoldPHIArgGEPIntoPHI(PHINode &PN); Instruction *FoldPHIArgLoadIntoPHI(PHINode &PN); - + Instruction *OptAndOp(Instruction *Op, ConstantInt *OpRHS, ConstantInt *AndRHS, BinaryOperator &TheAnd); - + Value *FoldLogicalPlusAnd(Value *LHS, Value *RHS, ConstantInt *Mask, bool isSub, Instruction &I); Value *InsertRangeTest(Value *V, Constant *Lo, Constant *Hi, @@ -382,8 +382,8 @@ private: Value *Descale(Value *Val, APInt Scale, bool &NoSignedWrap); }; - - + + } // end namespace llvm. #endif diff --git a/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/lib/Transforms/InstCombine/InstCombineAddSub.cpp index f07c58d7d0..03be8ef6fb 100644 --- a/lib/Transforms/InstCombine/InstCombineAddSub.cpp +++ b/lib/Transforms/InstCombine/InstCombineAddSub.cpp @@ -66,10 +66,12 @@ namespace { bool insaneIntVal(int V) { return V > 4 || V < -4; } APFloat *getFpValPtr(void) { return reinterpret_cast<APFloat*>(&FpValBuf.buffer[0]); } + const APFloat *getFpValPtr(void) const + { return reinterpret_cast<const APFloat*>(&FpValBuf.buffer[0]); } const APFloat &getFpVal(void) const { assert(IsFp && BufHasFpVal && "Incorret state"); - return *reinterpret_cast<const APFloat*>(&FpValBuf.buffer[0]); + return *getFpValPtr(); } APFloat &getFpVal(void) diff --git a/lib/Transforms/InstCombine/InstCombineCasts.cpp b/lib/Transforms/InstCombine/InstCombineCasts.cpp index c782032c45..0c0864f2c0 100644 --- a/lib/Transforms/InstCombine/InstCombineCasts.cpp +++ b/lib/Transforms/InstCombine/InstCombineCasts.cpp @@ -739,7 +739,7 @@ static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear) { } Instruction *InstCombiner::visitZExt(ZExtInst &CI) { - // If this zero extend is only used by a truncate, let the truncate by + // 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<TruncInst>(CI.use_back())) return 0; diff --git a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp index d0f43928c3..8e4267f898 100644 --- a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp +++ b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp @@ -377,6 +377,8 @@ Instruction *InstCombiner::visitFMul(BinaryOperator &I) { if (Value *V = SimplifyFMulInst(Op0, Op1, I.getFastMathFlags(), TD)) return ReplaceInstUsesWith(I, V); + bool AllowReassociate = I.hasUnsafeAlgebra(); + // Simplify mul instructions with a constant RHS. if (isa<Constant>(Op1)) { // Try to fold constant mul into select arguments. @@ -389,7 +391,7 @@ Instruction *InstCombiner::visitFMul(BinaryOperator &I) { return NV; ConstantFP *C = dyn_cast<ConstantFP>(Op1); - if (C && I.hasUnsafeAlgebra() && C->getValueAPF().isNormal()) { + if (C && AllowReassociate && C->getValueAPF().isNormal()) { // Let MDC denote an expression in one of these forms: // X * C, C/X, X/C, where C is a constant. // @@ -430,7 +432,7 @@ Instruction *InstCombiner::visitFMul(BinaryOperator &I) { BinaryOperator::CreateFAdd(M0, M1) : BinaryOperator::CreateFSub(M0, M1); Instruction *RI = cast<Instruction>(R); - RI->setHasUnsafeAlgebra(true); + RI->copyFastMathFlags(&I); return RI; } } @@ -438,9 +440,6 @@ Instruction *InstCombiner::visitFMul(BinaryOperator &I) { } } - if (Value *Op0v = dyn_castFNegVal(Op0)) // -X * -Y = X*Y - if (Value *Op1v = dyn_castFNegVal(Op1)) - return BinaryOperator::CreateFMul(Op0v, Op1v); // Under unsafe algebra do: // X * log2(0.5*Y) = X*log2(Y) - X @@ -469,36 +468,66 @@ Instruction *InstCombiner::visitFMul(BinaryOperator &I) { } } - // X * cond ? 1.0 : 0.0 => cond ? X : 0.0 - if (I.hasNoNaNs() && I.hasNoSignedZeros()) { - Value *V0 = I.getOperand(0); - Value *V1 = I.getOperand(1); - Value *Cond, *SLHS, *SRHS; - bool Match = false; - - if (match(V0, m_Select(m_Value(Cond), m_Value(SLHS), m_Value(SRHS)))) { - Match = true; - } else if (match(V1, m_Select(m_Value(Cond), m_Value(SLHS), - m_Value(SRHS)))) { - Match = true; - std::swap(V0, V1); + // Handle symmetric situation in a 2-iteration loop + Value *Opnd0 = Op0; + Value *Opnd1 = Op1; + for (int i = 0; i < 2; i++) { + bool IgnoreZeroSign = I.hasNoSignedZeros(); + if (BinaryOperator::isFNeg(Opnd0, IgnoreZeroSign)) { + Value *N0 = dyn_castFNegVal(Opnd0, IgnoreZeroSign); + Value *N1 = dyn_castFNegVal(Opnd1, IgnoreZeroSign); + + // -X * -Y => X*Y + if (N1) + return BinaryOperator::CreateFMul(N0, N1); + + if (Opnd0->hasOneUse()) { + // -X * Y => -(X*Y) (Promote negation as high as possible) + Value *T = Builder->CreateFMul(N0, Opnd1); + cast<Instruction>(T)->setDebugLoc(I.getDebugLoc()); + Instruction *Neg = BinaryOperator::CreateFNeg(T); + if (I.getFastMathFlags().any()) { + cast<Instruction>(T)->copyFastMathFlags(&I); + Neg->copyFastMathFlags(&I); + } + return Neg; + } } - if (Match) { - ConstantFP *C0 = dyn_cast<ConstantFP>(SLHS); - ConstantFP *C1 = dyn_cast<ConstantFP>(SRHS); - - if (C0 && C1 && - ((C0->isZero() && C1->isExactlyValue(1.0)) || - (C1->isZero() && C0->isExactlyValue(1.0)))) { - Value *T; - if (C0->isZero()) - T = Builder->CreateSelect(Cond, SLHS, V1); - else - T = Builder->CreateSelect(Cond, V1, SRHS); - return ReplaceInstUsesWith(I, T); + // (X*Y) * X => (X*X) * Y where Y != X + // The purpose is two-fold: + // 1) to form a power expression (of X). + // 2) potentially shorten the critical path: After transformation, the + // latency of the instruction Y is amortized by the expression of X*X, + // and therefore Y is in a "less critical" position compared to what it + // was before the transformation. + // + if (AllowReassociate) { + Value *Opnd0_0, *Opnd0_1; + if (Opnd0->hasOneUse() && + match(Opnd0, m_FMul(m_Value(Opnd0_0), m_Value(Opnd0_1)))) { + Value *Y = 0; + if (Opnd0_0 == Opnd1 && Opnd0_1 != Opnd1) + Y = Opnd0_1; + else if (Opnd0_1 == Opnd1 && Opnd0_0 != Opnd1) + Y = Opnd0_0; + + if (Y) { + Instruction *T = cast<Instruction>(Builder->CreateFMul(Opnd1, Opnd1)); + T->copyFastMathFlags(&I); + T->setDebugLoc(I.getDebugLoc()); + + Instruction *R = BinaryOperator::CreateFMul(T, Y); + R->copyFastMathFlags(&I); + return R; + } } } + + if (!isa<Constant>(Op1)) + std::swap(Opnd0, Opnd1); + else + break; } return Changed ? &I : 0; @@ -784,21 +813,140 @@ Instruction *InstCombiner::visitSDiv(BinaryOperator &I) { return 0; } +/// CvtFDivConstToReciprocal tries to convert X/C into X*1/C if C not a special +/// FP value and: +/// 1) 1/C is exact, or +/// 2) reciprocal is allowed. +/// If the convertion was successful, the simplified expression "X * 1/C" is +/// returned; otherwise, NULL is returned. +/// +static Instruction *CvtFDivConstToReciprocal(Value *Dividend, + ConstantFP *Divisor, + bool AllowReciprocal) { + const APFloat &FpVal = Divisor->getValueAPF(); + APFloat Reciprocal(FpVal.getSemantics()); + bool Cvt = FpVal.getExactInverse(&Reciprocal); + + if (!Cvt && AllowReciprocal && FpVal.isNormal()) { + Reciprocal = APFloat(FpVal.getSemantics(), 1.0f); + (void)Reciprocal.divide(FpVal, APFloat::rmNearestTiesToEven); + Cvt = !Reciprocal.isDenormal(); + } + + if (!Cvt) + return 0; + + ConstantFP *R; + R = ConstantFP::get(Dividend->getType()->getContext(), Reciprocal); + return BinaryOperator::CreateFMul(Dividend, R); +} + Instruction *InstCombiner::visitFDiv(BinaryOperator &I) { Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1); if (Value *V = SimplifyFDivInst(Op0, Op1, TD)) return ReplaceInstUsesWith(I, V); + bool AllowReassociate = I.hasUnsafeAlgebra(); + bool AllowReciprocal = I.hasAllowReciprocal(); + if (ConstantFP *Op1C = dyn_cast<ConstantFP>(Op1)) { - const APFloat &Op1F = Op1C->getValueAPF(); - - // If the divisor has an exact multiplicative inverse we can turn the fdiv - // into a cheaper fmul. - APFloat Reciprocal(Op1F.getSemantics()); - if (Op1F.getExactInverse(&Reciprocal)) { - ConstantFP *RFP = ConstantFP::get(Builder->getContext(), Reciprocal); - return BinaryOperator::CreateFMul(Op0, RFP); + if (AllowReassociate) { + ConstantFP *C1 = 0; + ConstantFP *C2 = Op1C; + Value *X; + Instruction *Res = 0; + + if (match(Op0, m_FMul(m_Value(X), m_ConstantFP(C1)))) { + // (X*C1)/C2 => X * (C1/C2) + // + Constant *C = ConstantExpr::getFDiv(C1, C2); + const APFloat &F = cast<ConstantFP>(C)->getValueAPF(); + if (F.isNormal() && !F.isDenormal()) + Res = BinaryOperator::CreateFMul(X, C); + } else if (match(Op0, m_FDiv(m_Value(X), m_ConstantFP(C1)))) { + // (X/C1)/C2 => X /(C2*C1) [=> X * 1/(C2*C1) if reciprocal is allowed] + // + Constant *C = ConstantExpr::getFMul(C1, C2); + const APFloat &F = cast<ConstantFP>(C)->getValueAPF(); + if (F.isNormal() && !F.isDenormal()) { + Res = CvtFDivConstToReciprocal(X, cast<ConstantFP>(C), + AllowReciprocal); + if (!Res) + Res = BinaryOperator::CreateFDiv(X, C); + } + } + + if (Res) { + Res->setFastMathFlags(I.getFastMathFlags()); + return Res; + } + } + + // X / C => X * 1/C + if (Instruction *T = CvtFDivConstToReciprocal(Op0, Op1C, AllowReciprocal)) + return T; + + return 0; + } + + if (AllowReassociate && isa<ConstantFP>(Op0)) { + ConstantFP *C1 = cast<ConstantFP>(Op0), *C2; + Constant *Fold = 0; + Value *X; + bool CreateDiv = true; + + // C1 / (X*C2) => (C1/C2) / X + if (match(Op1, m_FMul(m_Value(X), m_ConstantFP(C2)))) + Fold = ConstantExpr::getFDiv(C1, C2); + else if (match(Op1, m_FDiv(m_Value(X), m_ConstantFP(C2)))) { + // C1 / (X/C2) => (C1*C2) / X + Fold = ConstantExpr::getFMul(C1, C2); + } else if (match(Op1, m_FDiv(m_ConstantFP(C2), m_Value(X)))) { + // C1 / (C2/X) => (C1/C2) * X + Fold = ConstantExpr::getFDiv(C1, C2); + CreateDiv = false; + } + + if (Fold) { + const APFloat &FoldC = cast<ConstantFP>(Fold)->getValueAPF(); + if (FoldC.isNormal() && !FoldC.isDenormal()) { + Instruction *R = CreateDiv ? + BinaryOperator::CreateFDiv(Fold, X) : + BinaryOperator::CreateFMul(X, Fold); + R->setFastMathFlags(I.getFastMathFlags()); + return R; + } + } + return 0; + } + + if (AllowReassociate) { + Value *X, *Y; + Value *NewInst = 0; + Instruction *SimpR = 0; + + if (Op0->hasOneUse() && match(Op0, m_FDiv(m_Value(X), m_Value(Y)))) { + // (X/Y) / Z => X / (Y*Z) + // + if (!isa<ConstantFP>(Y) || !isa<ConstantFP>(Op1)) { + NewInst = Builder->CreateFMul(Y, Op1); + SimpR = BinaryOperator::CreateFDiv(X, NewInst); + } + } else if (Op1->hasOneUse() && match(Op1, m_FDiv(m_Value(X), m_Value(Y)))) { + // Z / (X/Y) => Z*Y / X + // + if (!isa<ConstantFP>(Y) || !isa<ConstantFP>(Op0)) { + NewInst = Builder->CreateFMul(Op0, Y); + SimpR = BinaryOperator::CreateFDiv(NewInst, X); + } + } + + if (NewInst) { + if (Instruction *T = dyn_cast<Instruction>(NewInst)) + T->setDebugLoc(I.getDebugLoc()); + SimpR->setFastMathFlags(I.getFastMathFlags()); + return SimpR; } } diff --git a/lib/Transforms/InstCombine/InstCombineWorklist.h b/lib/Transforms/InstCombine/InstCombineWorklist.h index 57ed9e32bb..49efce5c4f 100644 --- a/lib/Transforms/InstCombine/InstCombineWorklist.h +++ b/lib/Transforms/InstCombine/InstCombineWorklist.h @@ -19,20 +19,20 @@ #include "llvm/Support/raw_ostream.h" namespace llvm { - + /// InstCombineWorklist - This is the worklist management logic for /// InstCombine. class LLVM_LIBRARY_VISIBILITY InstCombineWorklist { SmallVector<Instruction*, 256> Worklist; DenseMap<Instruction*, unsigned> WorklistMap; - + void operator=(const InstCombineWorklist&RHS) LLVM_DELETED_FUNCTION; InstCombineWorklist(const InstCombineWorklist&) LLVM_DELETED_FUNCTION; public: InstCombineWorklist() {} - + bool isEmpty() const { return Worklist.empty(); } - + /// Add - Add the specified instruction to the worklist if it isn't already /// in it. void Add(Instruction *I) { @@ -41,12 +41,12 @@ public: Worklist.push_back(I); } } - + void AddValue(Value *V) { if (Instruction *I = dyn_cast<Instruction>(V)) Add(I); } - + /// AddInitialGroup - Add the specified batch of stuff in reverse order. /// which should only be done when the worklist is empty and when the group /// has no duplicates. @@ -61,25 +61,25 @@ public: Worklist.push_back(I); } } - + // Remove - remove I from the worklist if it exists. void Remove(Instruction *I) { DenseMap<Instruction*, unsigned>::iterator It = WorklistMap.find(I); if (It == WorklistMap.end()) return; // Not in worklist. - + // Don't bother moving everything down, just null out the slot. Worklist[It->second] = 0; - + WorklistMap.erase(It); } - + Instruction *RemoveOne() { Instruction *I = Worklist.back(); Worklist.pop_back(); WorklistMap.erase(I); return I; } - + /// AddUsersToWorkList - When an instruction is simplified, add all users of /// the instruction to the work lists because they might get more simplified /// now. @@ -89,18 +89,18 @@ public: UI != UE; ++UI) Add(cast<Instruction>(*UI)); } - - + + /// Zap - check that the worklist is empty and nuke the backing store for /// the map if it is large. void Zap() { assert(WorklistMap.empty() && "Worklist empty, but map not?"); - + // Do an explicit clear, this shrinks the map if needed. WorklistMap.clear(); } }; - + } // end namespace llvm. #endif diff --git a/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/lib/Transforms/Instrumentation/MemorySanitizer.cpp index 76da970682..db0de4d797 100644 --- a/lib/Transforms/Instrumentation/MemorySanitizer.cpp +++ b/lib/Transforms/Instrumentation/MemorySanitizer.cpp @@ -574,7 +574,7 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> { if (IntegerType *IT = dyn_cast<IntegerType>(OrigTy)) return IT; if (VectorType *VT = dyn_cast<VectorType>(OrigTy)) { - uint32_t EltSize = MS.TD->getTypeStoreSizeInBits(VT->getElementType()); + uint32_t EltSize = MS.TD->getTypeSizeInBits(VT->getElementType()); return VectorType::get(IntegerType::get(*MS.C, EltSize), VT->getNumElements()); } @@ -586,7 +586,7 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> { DEBUG(dbgs() << "getShadowTy: " << *ST << " ===> " << *Res << "\n"); return Res; } - uint32_t TypeSize = MS.TD->getTypeStoreSizeInBits(OrigTy); + uint32_t TypeSize = MS.TD->getTypeSizeInBits(OrigTy); return IntegerType::get(*MS.C, TypeSize); } @@ -1127,10 +1127,13 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> { Value *B = I.getOperand(1); Value *Sa = getShadow(A); Value *Sb = getShadow(B); - if (A->getType()->isPointerTy()) - A = IRB.CreatePointerCast(A, MS.IntptrTy); - if (B->getType()->isPointerTy()) - B = IRB.CreatePointerCast(B, MS.IntptrTy); + + // Get rid of pointers and vectors of pointers. + // For ints (and vectors of ints), types of A and Sa match, + // and this is a no-op. + A = IRB.CreatePointerCast(A, Sa->getType()); + B = IRB.CreatePointerCast(B, Sb->getType()); + // A == B <==> (C = A^B) == 0 // A != B <==> (C = A^B) != 0 // Sc = Sa | Sb diff --git a/lib/Transforms/Utils/BasicBlockUtils.cpp b/lib/Transforms/Utils/BasicBlockUtils.cpp index 8330e8468f..ba99d2e662 100644 --- a/lib/Transforms/Utils/BasicBlockUtils.cpp +++ b/lib/Transforms/Utils/BasicBlockUtils.cpp @@ -37,12 +37,12 @@ void llvm::DeleteDeadBlock(BasicBlock *BB) { // Can delete self loop. BB->getSinglePredecessor() == BB) && "Block is not dead!"); TerminatorInst *BBTerm = BB->getTerminator(); - + // Loop through all of our successors and make sure they know that one // of their predecessors is going away. for (unsigned i = 0, e = BBTerm->getNumSuccessors(); i != e; ++i) BBTerm->getSuccessor(i)->removePredecessor(BB); - + // Zap all the instructions in the block. while (!BB->empty()) { Instruction &I = BB->back(); @@ -55,7 +55,7 @@ void llvm::DeleteDeadBlock(BasicBlock *BB) { I.replaceAllUsesWith(UndefValue::get(I.getType())); BB->getInstList().pop_back(); } - + // Zap the block! BB->eraseFromParent(); } @@ -66,25 +66,25 @@ void llvm::DeleteDeadBlock(BasicBlock *BB) { /// when the block has exactly one predecessor. void llvm::FoldSingleEntryPHINodes(BasicBlock *BB, Pass *P) { if (!isa<PHINode>(BB->begin())) return; - + AliasAnalysis *AA = 0; MemoryDependenceAnalysis *MemDep = 0; if (P) { AA = P->getAnalysisIfAvailable<AliasAnalysis>(); MemDep = P->getAnalysisIfAvailable<MemoryDependenceAnalysis>(); } - + while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) { if (PN->getIncomingValue(0) != PN) PN->replaceAllUsesWith(PN->getIncomingValue(0)); else PN->replaceAllUsesWith(UndefValue::get(PN->getType())); - + if (MemDep) MemDep->removeInstruction(PN); // Memdep updates AA itself. else if (AA && isa<PointerType>(PN->getType())) AA->deleteValue(PN); - + PN->eraseFromParent(); } } @@ -115,7 +115,7 @@ bool llvm::DeleteDeadPHIs(BasicBlock *BB, const TargetLibraryInfo *TLI) { bool llvm::MergeBlockIntoPredecessor(BasicBlock *BB, Pass *P) { // Don't merge away blocks who have their address taken. if (BB->hasAddressTaken()) return false; - + // Can't merge if there are multiple predecessors, or no predecessors. BasicBlock *PredBB = BB->getUniquePredecessor(); if (!PredBB) return false; @@ -124,7 +124,7 @@ bool llvm::MergeBlockIntoPredecessor(BasicBlock *BB, Pass *P) { if (PredBB == BB) return false; // Don't break invokes. if (isa<InvokeInst>(PredBB->getTerminator())) return false; - + succ_iterator SI(succ_begin(PredBB)), SE(succ_end(PredBB)); BasicBlock *OnlySucc = BB; for (; SI != SE; ++SI) @@ -132,7 +132,7 @@ bool llvm::MergeBlockIntoPredecessor(BasicBlock *BB, Pass *P) { OnlySucc = 0; // There are multiple distinct successors! break; } - + // Can't merge if there are multiple successors. if (!OnlySucc) return false; @@ -149,21 +149,21 @@ bool llvm::MergeBlockIntoPredecessor(BasicBlock *BB, Pass *P) { // Begin by getting rid of unneeded PHIs. if (isa<PHINode>(BB->front())) FoldSingleEntryPHINodes(BB, P); - + // Delete the unconditional branch from the predecessor... PredBB->getInstList().pop_back(); - + // Make all PHI nodes that referred to BB now refer to Pred as their // source... BB->replaceAllUsesWith(PredBB); - + // Move all definitions in the successor to the predecessor... PredBB->getInstList().splice(PredBB->end(), BB->getInstList()); - + // Inherit predecessors name if it exists. if (!PredBB->hasName()) PredBB->takeName(BB); - + // Finally, erase the old block and update dominator info. if (P) { if (DominatorTree *DT = P->getAnalysisIfAvailable<DominatorTree>()) { @@ -176,16 +176,16 @@ bool llvm::MergeBlockIntoPredecessor(BasicBlock *BB, Pass *P) { DT->eraseNode(BB); } - + if (LoopInfo *LI = P->getAnalysisIfAvailable<LoopInfo>()) LI->removeBlock(BB); - + if (MemoryDependenceAnalysis *MD = P->getAnalysisIfAvailable<MemoryDependenceAnalysis>()) MD->invalidateCachedPredecessors(); } } - + BB->eraseFromParent(); return true; } @@ -251,11 +251,11 @@ unsigned llvm::GetSuccessorNumber(BasicBlock *BB, BasicBlock *Succ) { } } -/// SplitEdge - Split the edge connecting specified block. Pass P must -/// not be NULL. +/// SplitEdge - Split the edge connecting specified block. Pass P must +/// not be NULL. BasicBlock *llvm::SplitEdge(BasicBlock *BB, BasicBlock *Succ, Pass *P) { unsigned SuccNum = GetSuccessorNumber(BB, Succ); - + // If this is a critical edge, let SplitCriticalEdge do it. TerminatorInst *LatchTerm = BB->getTerminator(); if (SplitCriticalEdge(LatchTerm, SuccNum, P)) @@ -271,11 +271,11 @@ BasicBlock *llvm::SplitEdge(BasicBlock *BB, BasicBlock *Succ, Pass *P) { SP = NULL; return SplitBlock(Succ, Succ->begin(), P); } - + // Otherwise, if BB has a single successor, split it at the bottom of the // block. assert(BB->getTerminator()->getNumSuccessors() == 1 && - "Should have a single succ!"); + "Should have a single succ!"); return SplitBlock(BB, BB->getTerminator(), P); } @@ -301,12 +301,12 @@ BasicBlock *llvm::SplitBlock(BasicBlock *Old, Instruction *SplitPt, Pass *P) { if (DomTreeNode *OldNode = DT->getNode(Old)) { std::vector<DomTreeNode *> Children; for (DomTreeNode::iterator I = OldNode->begin(), E = OldNode->end(); - I != E; ++I) + I != E; ++I) Children.push_back(*I); DomTreeNode *NewNode = DT->addNewBlock(New,Old); for (std::vector<DomTreeNode *>::iterator I = Children.begin(), - E = Children.end(); I != E; ++I) + E = Children.end(); I != E; ++I) DT->changeImmediateDominator(*I, NewNode); } } @@ -424,7 +424,7 @@ static void UpdatePHINodes(BasicBlock *OrigBB, BasicBlock *NewBB, PHINode *NewPHI = PHINode::Create(PN->getType(), Preds.size(), PN->getName() + ".ph", BI); if (AA) AA->copyValue(PN, NewPHI); - + // Move all of the PHI values for 'Preds' to the new PHI. for (unsigned i = 0, e = Preds.size(); i != e; ++i) { Value *V = PN->removeIncomingValue(Preds[i], false); @@ -451,16 +451,16 @@ static void UpdatePHINodes(BasicBlock *OrigBB, BasicBlock *NewBB, /// preserve LoopSimplify (because it's complicated to handle the case where one /// of the edges being split is an exit of a loop with other exits). /// -BasicBlock *llvm::SplitBlockPredecessors(BasicBlock *BB, +BasicBlock *llvm::SplitBlockPredecessors(BasicBlock *BB, ArrayRef<BasicBlock*> Preds, const char *Suffix, Pass *P) { // Create new basic block, insert right before the original block. BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), BB->getName()+Suffix, BB->getParent(), BB); - + // The new block unconditionally branches to the old block. BranchInst *BI = BranchInst::Create(BB, NewBB); - + // Move the edges from Preds to point to NewBB instead of BB. for (unsigned i = 0, e = Preds.size(); i != e; ++i) { // This is slightly more strict than necessary; the minimum requirement @@ -497,13 +497,13 @@ BasicBlock *llvm::SplitBlockPredecessors(BasicBlock *BB, /// block gets the remaining predecessors of OrigBB. The landingpad instruction /// OrigBB is clone into both of the new basic blocks. The new blocks are given /// the suffixes 'Suffix1' and 'Suffix2', and are returned in the NewBBs vector. -/// +/// /// This currently updates the LLVM IR, AliasAnalysis, DominatorTree, /// DominanceFrontier, LoopInfo, and LCCSA but no other analyses. In particular, /// it does not preserve LoopSimplify (because it's complicated to handle the /// case where one of the edges being split is an exit of a loop with other /// exits). -/// +/// void llvm::SplitLandingPadPredecessors(BasicBlock *OrigBB, ArrayRef<BasicBlock*> Preds, const char *Suffix1, const char *Suffix2, @@ -608,11 +608,11 @@ void llvm::FindFunctionBackedges(const Function &F, const BasicBlock *BB = &F.getEntryBlock(); if (succ_begin(BB) == succ_end(BB)) return; - + SmallPtrSet<const BasicBlock*, 8> Visited; SmallVector<std::pair<const BasicBlock*, succ_const_iterator>, 8> VisitStack; SmallPtrSet<const BasicBlock*, 8> InStack; - + Visited.insert(BB); VisitStack.push_back(std::make_pair(BB, succ_begin(BB))); InStack.insert(BB); @@ -620,7 +620,7 @@ void llvm::FindFunctionBackedges(const Function &F, std::pair<const BasicBlock*, succ_const_iterator> &Top = VisitStack.back(); const BasicBlock *ParentBB = Top.first; succ_const_iterator &I = Top.second; - + bool FoundNew = false; while (I != succ_end(ParentBB)) { BB = *I++; @@ -632,7 +632,7 @@ void llvm::FindFunctionBackedges(const Function &F, if (InStack.count(BB)) Result.push_back(std::make_pair(ParentBB, BB)); } - + if (FoundNew) { // Go down one level if there is a unvisited successor. InStack.insert(BB); @@ -641,7 +641,7 @@ void llvm::FindFunctionBackedges(const Function &F, // Go up one level. InStack.erase(VisitStack.pop_back_val().first); } - } while (!VisitStack.empty()); + } while (!VisitStack.empty()); } /// FoldReturnIntoUncondBranch - This method duplicates the specified return @@ -655,7 +655,7 @@ ReturnInst *llvm::FoldReturnIntoUncondBranch(ReturnInst *RI, BasicBlock *BB, // Clone the return and add it to the end of the predecessor. Instruction *NewRet = RI->clone(); Pred->getInstList().push_back(NewRet); - + // If the return instruction returns a value, and if the value was a // PHI node in "BB", propagate the right value into the return. for (User::op_iterator i = NewRet->op_begin(), e = NewRet->op_end(); @@ -679,7 +679,7 @@ ReturnInst *llvm::FoldReturnIntoUncondBranch(ReturnInst *RI, BasicBlock *BB, } } } - + // Update any PHI nodes in the returning block to realize that we no // longer branch to them. BB->removePredecessor(Pred); diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp index 464ed97506..bc8e1217be 100644 --- a/lib/Transforms/Vectorize/LoopVectorize.cpp +++ b/lib/Transforms/Vectorize/LoopVectorize.cpp @@ -2633,7 +2633,7 @@ LoopVectorizationCostModel::selectVectorizationFactor(bool OptForSize, if (MaxVectorSize == 0) { DEBUG(dbgs() << "LV: The target has no vector registers.\n"); - return 1; + MaxVectorSize = 1; } assert(MaxVectorSize <= 32 && "Did not expect to pack so many elements" |