From 8e0d1c03ca7fd86e6879b4e37d0d7f0e982feef6 Mon Sep 17 00:00:00 2001 From: Benjamin Kramer Date: Wed, 29 Aug 2012 15:32:21 +0000 Subject: Make MemoryBuiltins aware of TargetLibraryInfo. This disables malloc-specific optimization when -fno-builtin (or -ffreestanding) is specified. This has been a problem for a long time but became more severe with the recent memory builtin improvements. Since the memory builtin functions are used everywhere, this required passing TLI in many places. This means that functions that now have an optional TLI argument, like RecursivelyDeleteTriviallyDeadFunctions, won't remove dead mallocs anymore if the TLI argument is missing. I've updated most passes to do the right thing. Fixes PR13694 and probably others. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@162841 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/llvm/Analysis/AliasAnalysis.h | 9 +- include/llvm/Analysis/MemoryBuiltins.h | 58 +++++---- include/llvm/Target/TargetLibraryInfo.h | 38 ++++++ include/llvm/Transforms/Utils/BasicBlockUtils.h | 3 +- include/llvm/Transforms/Utils/Local.h | 14 +- lib/Analysis/AliasAnalysis.cpp | 2 + lib/Analysis/BasicAliasAnalysis.cpp | 20 +-- lib/Analysis/IPA/GlobalsModRef.cpp | 6 +- lib/Analysis/MemoryBuiltins.cpp | 142 +++++++++++++-------- lib/Analysis/MemoryDependenceAnalysis.cpp | 4 +- lib/Target/TargetLibraryInfo.cpp | 21 ++- lib/Transforms/IPO/GlobalOpt.cpp | 33 ++--- lib/Transforms/IPO/Inliner.cpp | 4 +- lib/Transforms/InstCombine/InstCombineCalls.cpp | 6 +- .../InstCombine/InstructionCombining.cpp | 15 ++- lib/Transforms/Instrumentation/BoundsChecking.cpp | 6 +- lib/Transforms/Scalar/CodeGenPrepare.cpp | 2 +- lib/Transforms/Scalar/DCE.cpp | 8 +- lib/Transforms/Scalar/DeadStoreElimination.cpp | 26 ++-- lib/Transforms/Scalar/EarlyCSE.cpp | 2 +- lib/Transforms/Scalar/GVN.cpp | 4 +- lib/Transforms/Scalar/IndVarSimplify.cpp | 17 ++- lib/Transforms/Scalar/JumpThreading.cpp | 2 +- lib/Transforms/Scalar/LICM.cpp | 2 +- lib/Transforms/Scalar/LoopIdiomRecognize.cpp | 24 ++-- lib/Transforms/Scalar/LoopInstSimplify.cpp | 2 +- lib/Transforms/Utils/BasicBlockUtils.cpp | 4 +- lib/Transforms/Utils/Local.cpp | 36 +++--- lib/Transforms/Utils/SimplifyInstructions.cpp | 2 +- lib/Transforms/Vectorize/BBVectorize.cpp | 2 +- test/Transforms/GVN/malloc-load-removal.ll | 31 +++++ 31 files changed, 361 insertions(+), 184 deletions(-) create mode 100644 test/Transforms/GVN/malloc-load-removal.ll diff --git a/include/llvm/Analysis/AliasAnalysis.h b/include/llvm/Analysis/AliasAnalysis.h index 674868a..6c936f0 100644 --- a/include/llvm/Analysis/AliasAnalysis.h +++ b/include/llvm/Analysis/AliasAnalysis.h @@ -46,6 +46,7 @@ class LoadInst; class StoreInst; class VAArgInst; class TargetData; +class TargetLibraryInfo; class Pass; class AnalysisUsage; class MemTransferInst; @@ -55,6 +56,7 @@ class DominatorTree; class AliasAnalysis { protected: const TargetData *TD; + const TargetLibraryInfo *TLI; private: AliasAnalysis *AA; // Previous Alias Analysis to chain to. @@ -73,7 +75,7 @@ protected: public: static char ID; // Class identification, replacement for typeinfo - AliasAnalysis() : TD(0), AA(0) {} + AliasAnalysis() : TD(0), TLI(0), AA(0) {} virtual ~AliasAnalysis(); // We want to be subclassed /// UnknownSize - This is a special value which can be used with the @@ -86,6 +88,11 @@ public: /// const TargetData *getTargetData() const { return TD; } + /// getTargetLibraryInfo - Return a pointer to the current TargetLibraryInfo + /// object, or null if no TargetLibraryInfo object is available. + /// + const TargetLibraryInfo *getTargetLibraryInfo() const { return TLI; } + /// getTypeStoreSize - Return the TargetData store size for the given type, /// if known, or a conservative value otherwise. /// diff --git a/include/llvm/Analysis/MemoryBuiltins.h b/include/llvm/Analysis/MemoryBuiltins.h index e16f389..c3ae603 100644 --- a/include/llvm/Analysis/MemoryBuiltins.h +++ b/include/llvm/Analysis/MemoryBuiltins.h @@ -28,6 +28,7 @@ namespace llvm { class CallInst; class PointerType; class TargetData; +class TargetLibraryInfo; class Type; class Value; @@ -35,27 +36,33 @@ class Value; /// \brief Tests if a value is a call or invoke to a library function that /// allocates or reallocates memory (either malloc, calloc, realloc, or strdup /// like). -bool isAllocationFn(const Value *V, bool LookThroughBitCast = false); +bool isAllocationFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast = false); /// \brief Tests if a value is a call or invoke to a function that returns a /// NoAlias pointer (including malloc/calloc/realloc/strdup-like functions). -bool isNoAliasFn(const Value *V, bool LookThroughBitCast = false); +bool isNoAliasFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast = false); /// \brief Tests if a value is a call or invoke to a library function that /// allocates uninitialized memory (such as malloc). -bool isMallocLikeFn(const Value *V, bool LookThroughBitCast = false); +bool isMallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast = false); /// \brief Tests if a value is a call or invoke to a library function that /// allocates zero-filled memory (such as calloc). -bool isCallocLikeFn(const Value *V, bool LookThroughBitCast = false); +bool isCallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast = false); /// \brief Tests if a value is a call or invoke to a library function that /// allocates memory (either malloc, calloc, or strdup like). -bool isAllocLikeFn(const Value *V, bool LookThroughBitCast = false); +bool isAllocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast = false); /// \brief Tests if a value is a call or invoke to a library function that /// reallocates memory (such as realloc). -bool isReallocLikeFn(const Value *V, bool LookThroughBitCast = false); +bool isReallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast = false); //===----------------------------------------------------------------------===// @@ -65,29 +72,31 @@ bool isReallocLikeFn(const Value *V, bool LookThroughBitCast = false); /// extractMallocCall - Returns the corresponding CallInst if the instruction /// is a malloc call. Since CallInst::CreateMalloc() only creates calls, we /// ignore InvokeInst here. -const CallInst *extractMallocCall(const Value *I); -static inline CallInst *extractMallocCall(Value *I) { - return const_cast(extractMallocCall((const Value*)I)); +const CallInst *extractMallocCall(const Value *I, const TargetLibraryInfo *TLI); +static inline CallInst *extractMallocCall(Value *I, + const TargetLibraryInfo *TLI) { + return const_cast(extractMallocCall((const Value*)I, TLI)); } /// isArrayMalloc - Returns the corresponding CallInst if the instruction /// is a call to malloc whose array size can be determined and the array size /// is not constant 1. Otherwise, return NULL. -const CallInst *isArrayMalloc(const Value *I, const TargetData *TD); +const CallInst *isArrayMalloc(const Value *I, const TargetData *TD, + const TargetLibraryInfo *TLI); /// getMallocType - Returns the PointerType resulting from the malloc call. /// The PointerType depends on the number of bitcast uses of the malloc call: /// 0: PointerType is the malloc calls' return type. /// 1: PointerType is the bitcast's result type. /// >1: Unique PointerType cannot be determined, return NULL. -PointerType *getMallocType(const CallInst *CI); +PointerType *getMallocType(const CallInst *CI, const TargetLibraryInfo *TLI); /// getMallocAllocatedType - Returns the Type allocated by malloc call. /// The Type depends on the number of bitcast uses of the malloc call: /// 0: PointerType is the malloc calls' return type. /// 1: PointerType is the bitcast's result type. /// >1: Unique PointerType cannot be determined, return NULL. -Type *getMallocAllocatedType(const CallInst *CI); +Type *getMallocAllocatedType(const CallInst *CI, const TargetLibraryInfo *TLI); /// getMallocArraySize - Returns the array size of a malloc call. If the /// argument passed to malloc is a multiple of the size of the malloced type, @@ -95,6 +104,7 @@ Type *getMallocAllocatedType(const CallInst *CI); /// constant 1. Otherwise, return NULL for mallocs whose array size cannot be /// determined. Value *getMallocArraySize(CallInst *CI, const TargetData *TD, + const TargetLibraryInfo *TLI, bool LookThroughSExt = false); @@ -104,9 +114,10 @@ Value *getMallocArraySize(CallInst *CI, const TargetData *TD, /// extractCallocCall - Returns the corresponding CallInst if the instruction /// is a calloc call. -const CallInst *extractCallocCall(const Value *I); -static inline CallInst *extractCallocCall(Value *I) { - return const_cast(extractCallocCall((const Value*)I)); +const CallInst *extractCallocCall(const Value *I, const TargetLibraryInfo *TLI); +static inline CallInst *extractCallocCall(Value *I, + const TargetLibraryInfo *TLI) { + return const_cast(extractCallocCall((const Value*)I, TLI)); } @@ -115,10 +126,10 @@ static inline CallInst *extractCallocCall(Value *I) { // /// isFreeCall - Returns non-null if the value is a call to the builtin free() -const CallInst *isFreeCall(const Value *I); +const CallInst *isFreeCall(const Value *I, const TargetLibraryInfo *TLI); -static inline CallInst *isFreeCall(Value *I) { - return const_cast(isFreeCall((const Value*)I)); +static inline CallInst *isFreeCall(Value *I, const TargetLibraryInfo *TLI) { + return const_cast(isFreeCall((const Value*)I, TLI)); } @@ -131,7 +142,7 @@ static inline CallInst *isFreeCall(Value *I) { /// If RoundToAlign is true, then Size is rounded up to the aligment of allocas, /// byval arguments, and global variables. bool getObjectSize(const Value *Ptr, uint64_t &Size, const TargetData *TD, - bool RoundToAlign = false); + const TargetLibraryInfo *TLI, bool RoundToAlign = false); @@ -143,6 +154,7 @@ class ObjectSizeOffsetVisitor : public InstVisitor { const TargetData *TD; + const TargetLibraryInfo *TLI; bool RoundToAlign; unsigned IntTyBits; APInt Zero; @@ -155,8 +167,8 @@ class ObjectSizeOffsetVisitor } public: - ObjectSizeOffsetVisitor(const TargetData *TD, LLVMContext &Context, - bool RoundToAlign = false); + ObjectSizeOffsetVisitor(const TargetData *TD, const TargetLibraryInfo *TLI, + LLVMContext &Context, bool RoundToAlign = false); SizeOffsetType compute(Value *V); @@ -202,6 +214,7 @@ class ObjectSizeOffsetEvaluator typedef SmallPtrSet PtrSetTy; const TargetData *TD; + const TargetLibraryInfo *TLI; LLVMContext &Context; BuilderTy Builder; IntegerType *IntTy; @@ -215,7 +228,8 @@ class ObjectSizeOffsetEvaluator SizeOffsetEvalType compute_(Value *V); public: - ObjectSizeOffsetEvaluator(const TargetData *TD, LLVMContext &Context); + ObjectSizeOffsetEvaluator(const TargetData *TD, const TargetLibraryInfo *TLI, + LLVMContext &Context); SizeOffsetEvalType compute(Value *V); bool knownSize(SizeOffsetEvalType SizeOffset) { diff --git a/include/llvm/Target/TargetLibraryInfo.h b/include/llvm/Target/TargetLibraryInfo.h index a44105b..2a0a432 100644 --- a/include/llvm/Target/TargetLibraryInfo.h +++ b/include/llvm/Target/TargetLibraryInfo.h @@ -18,6 +18,26 @@ namespace llvm { namespace LibFunc { enum Func { + /// void operator delete[](void*); + ZdaPv, + /// void operator delete(void*); + ZdlPv, + /// void *new[](unsigned int); + Znaj, + /// void *new[](unsigned int, nothrow); + ZnajRKSt9nothrow_t, + /// void *new[](unsigned long); + Znam, + /// void *new[](unsigned long, nothrow); + ZnamRKSt9nothrow_t, + /// void *new(unsigned int); + Znwj, + /// void *new(unsigned int, nothrow); + ZnwjRKSt9nothrow_t, + /// void *new(unsigned long); + Znwm, + /// void *new(unsigned long, nothrow); + ZnwmRKSt9nothrow_t, /// int __cxa_atexit(void (*f)(void *), void *p, void *d); cxa_atexit, /// void __cxa_guard_abort(guard_t *guard); @@ -71,6 +91,8 @@ namespace llvm { atanhl, /// long double atanl(long double x); atanl, + /// void *calloc(size_t count, size_t size); + calloc, /// double cbrt(double x); cbrt, /// float cbrtf(float x); @@ -149,6 +171,8 @@ namespace llvm { fputc, /// int fputs(const char *s, FILE *stream); fputs, + /// void free(void *ptr); + free, /// size_t fwrite(const void *ptr, size_t size, size_t nitems, /// FILE *stream); fwrite, @@ -184,6 +208,8 @@ namespace llvm { logf, /// long double logl(long double x); logl, + /// void *malloc(size_t size); + malloc, /// void *memchr(const void *s, int c, size_t n); memchr, /// int memcmp(const void *s1, const void *s2, size_t n); @@ -202,6 +228,8 @@ namespace llvm { nearbyintf, /// long double nearbyintl(long double x); nearbyintl, + /// int posix_memalign(void **memptr, size_t alignment, size_t size); + posix_memalign, /// double pow(double x, double y); pow, /// float powf(float x, float y); @@ -212,6 +240,10 @@ namespace llvm { putchar, /// int puts(const char *s); puts, + /// void *realloc(void *ptr, size_t size); + realloc, + /// void *reallocf(void *ptr, size_t size); + reallocf, /// double rint(double x); rint, /// float rintf(float x); @@ -250,6 +282,8 @@ namespace llvm { strchr, /// char *strcpy(char *s1, const char *s2); strcpy, + /// char *strdup(const char *s1); + strdup, /// size_t strlen(const char *s); strlen, /// char *strncat(char *s1, const char *s2, size_t n); @@ -258,6 +292,8 @@ namespace llvm { strncmp, /// char *strncpy(char *s1, const char *s2, size_t n); strncpy, + /// char *strndup(const char *s1, size_t n); + strndup, /// size_t strnlen(const char *s, size_t maxlen); strnlen, /// double tan(double x); @@ -278,6 +314,8 @@ namespace llvm { truncf, /// long double truncl(long double x); truncl, + /// void *valloc(size_t size); + valloc, NumLibFuncs }; diff --git a/include/llvm/Transforms/Utils/BasicBlockUtils.h b/include/llvm/Transforms/Utils/BasicBlockUtils.h index 8a939cc..2510aec 100644 --- a/include/llvm/Transforms/Utils/BasicBlockUtils.h +++ b/include/llvm/Transforms/Utils/BasicBlockUtils.h @@ -27,6 +27,7 @@ class AliasAnalysis; class Instruction; class Pass; class ReturnInst; +class TargetLibraryInfo; /// DeleteDeadBlock - Delete the specified block, which must have no /// predecessors. @@ -44,7 +45,7 @@ void FoldSingleEntryPHINodes(BasicBlock *BB, Pass *P = 0); /// a result. This includes tracing the def-use list from the PHI to see if /// it is ultimately unused or if it reaches an unused cycle. Return true /// if any PHIs were deleted. -bool DeleteDeadPHIs(BasicBlock *BB); +bool DeleteDeadPHIs(BasicBlock *BB, const TargetLibraryInfo *TLI = 0); /// MergeBlockIntoPredecessor - Attempts to merge a block into its predecessor, /// if possible. The return value indicates success or failure. diff --git a/include/llvm/Transforms/Utils/Local.h b/include/llvm/Transforms/Utils/Local.h index 495eab7..83f0e7a 100644 --- a/include/llvm/Transforms/Utils/Local.h +++ b/include/llvm/Transforms/Utils/Local.h @@ -36,6 +36,7 @@ class PHINode; class AllocaInst; class ConstantExpr; class TargetData; +class TargetLibraryInfo; class DIBuilder; template class SmallVectorImpl; @@ -51,7 +52,8 @@ template class SmallVectorImpl; /// Also calls RecursivelyDeleteTriviallyDeadInstructions() on any branch/switch /// conditions and indirectbr addresses this might make dead if /// DeleteDeadConditions is true. -bool ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions = false); +bool ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions = false, + const TargetLibraryInfo *TLI = 0); //===----------------------------------------------------------------------===// // Local dead code elimination. @@ -60,20 +62,21 @@ bool ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions = false); /// isInstructionTriviallyDead - Return true if the result produced by the /// instruction is not used, and the instruction has no side effects. /// -bool isInstructionTriviallyDead(Instruction *I); +bool isInstructionTriviallyDead(Instruction *I, const TargetLibraryInfo *TLI=0); /// RecursivelyDeleteTriviallyDeadInstructions - If the specified value is a /// trivially dead instruction, delete it. If that makes any of its operands /// trivially dead, delete them too, recursively. Return true if any /// instructions were deleted. -bool RecursivelyDeleteTriviallyDeadInstructions(Value *V); +bool RecursivelyDeleteTriviallyDeadInstructions(Value *V, + const TargetLibraryInfo *TLI=0); /// RecursivelyDeleteDeadPHINode - If the specified value is an effectively /// dead PHI node, due to being a def-use chain of single-use nodes that /// either forms a cycle or is terminated by a trivially dead instruction, /// delete it. If that makes any of its operands trivially dead, delete them /// too, recursively. Return true if a change was made. -bool RecursivelyDeleteDeadPHINode(PHINode *PN); +bool RecursivelyDeleteDeadPHINode(PHINode *PN, const TargetLibraryInfo *TLI=0); /// SimplifyInstructionsInBlock - Scan the specified basic block and try to @@ -81,7 +84,8 @@ bool RecursivelyDeleteDeadPHINode(PHINode *PN); /// /// This returns true if it changed the code, note that it can delete /// instructions in other blocks as well in this block. -bool SimplifyInstructionsInBlock(BasicBlock *BB, const TargetData *TD = 0); +bool SimplifyInstructionsInBlock(BasicBlock *BB, const TargetData *TD = 0, + const TargetLibraryInfo *TLI = 0); //===----------------------------------------------------------------------===// // Control Flow Graph Restructuring. diff --git a/lib/Analysis/AliasAnalysis.cpp b/lib/Analysis/AliasAnalysis.cpp index 3b6aab1..f768eec 100644 --- a/lib/Analysis/AliasAnalysis.cpp +++ b/lib/Analysis/AliasAnalysis.cpp @@ -36,6 +36,7 @@ #include "llvm/LLVMContext.h" #include "llvm/Type.h" #include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetLibraryInfo.h" using namespace llvm; // Register the AliasAnalysis interface, providing a nice name to refer to. @@ -452,6 +453,7 @@ AliasAnalysis::~AliasAnalysis() {} /// void AliasAnalysis::InitializeAliasAnalysis(Pass *P) { TD = P->getAnalysisIfAvailable(); + TLI = P->getAnalysisIfAvailable(); AA = &P->getAnalysis(); } diff --git a/lib/Analysis/BasicAliasAnalysis.cpp b/lib/Analysis/BasicAliasAnalysis.cpp index 1d028c2..e3476ac 100644 --- a/lib/Analysis/BasicAliasAnalysis.cpp +++ b/lib/Analysis/BasicAliasAnalysis.cpp @@ -85,9 +85,10 @@ static bool isEscapeSource(const Value *V) { /// getObjectSize - Return the size of the object specified by V, or /// UnknownSize if unknown. static uint64_t getObjectSize(const Value *V, const TargetData &TD, + const TargetLibraryInfo &TLI, bool RoundToAlign = false) { uint64_t Size; - if (getObjectSize(V, Size, &TD, RoundToAlign)) + if (getObjectSize(V, Size, &TD, &TLI, RoundToAlign)) return Size; return AliasAnalysis::UnknownSize; } @@ -95,10 +96,11 @@ static uint64_t getObjectSize(const Value *V, const TargetData &TD, /// isObjectSmallerThan - Return true if we can prove that the object specified /// by V is smaller than Size. static bool isObjectSmallerThan(const Value *V, uint64_t Size, - const TargetData &TD) { + const TargetData &TD, + const TargetLibraryInfo &TLI) { // This function needs to use the aligned object size because we allow // reads a bit past the end given sufficient alignment. - uint64_t ObjectSize = getObjectSize(V, TD, /*RoundToAlign*/true); + uint64_t ObjectSize = getObjectSize(V, TD, TLI, /*RoundToAlign*/true); return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize < Size; } @@ -106,8 +108,8 @@ static bool isObjectSmallerThan(const Value *V, uint64_t Size, /// isObjectSize - Return true if we can prove that the object specified /// by V has size Size. static bool isObjectSize(const Value *V, uint64_t Size, - const TargetData &TD) { - uint64_t ObjectSize = getObjectSize(V, TD); + const TargetData &TD, const TargetLibraryInfo &TLI) { + uint64_t ObjectSize = getObjectSize(V, TD, TLI); return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize == Size; } @@ -1133,8 +1135,8 @@ BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size, // If the size of one access is larger than the entire object on the other // side, then we know such behavior is undefined and can assume no alias. if (TD) - if ((V1Size != UnknownSize && isObjectSmallerThan(O2, V1Size, *TD)) || - (V2Size != UnknownSize && isObjectSmallerThan(O1, V2Size, *TD))) + if ((V1Size != UnknownSize && isObjectSmallerThan(O2, V1Size, *TD, *TLI)) || + (V2Size != UnknownSize && isObjectSmallerThan(O1, V2Size, *TD, *TLI))) return NoAlias; // Check the cache before climbing up use-def chains. This also terminates @@ -1184,8 +1186,8 @@ BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size, // accesses is accessing the entire object, then the accesses must // overlap in some way. if (TD && O1 == O2) - if ((V1Size != UnknownSize && isObjectSize(O1, V1Size, *TD)) || - (V2Size != UnknownSize && isObjectSize(O2, V2Size, *TD))) + if ((V1Size != UnknownSize && isObjectSize(O1, V1Size, *TD, *TLI)) || + (V2Size != UnknownSize && isObjectSize(O2, V2Size, *TD, *TLI))) return AliasCache[Locs] = PartialAlias; AliasResult Result = diff --git a/lib/Analysis/IPA/GlobalsModRef.cpp b/lib/Analysis/IPA/GlobalsModRef.cpp index 22f6e96..990caa8 100644 --- a/lib/Analysis/IPA/GlobalsModRef.cpp +++ b/lib/Analysis/IPA/GlobalsModRef.cpp @@ -263,7 +263,7 @@ bool GlobalsModRef::AnalyzeUsesOfPointer(Value *V, } else if (BitCastInst *BCI = dyn_cast(U)) { if (AnalyzeUsesOfPointer(BCI, Readers, Writers, OkayStoreDest)) return true; - } else if (isFreeCall(U)) { + } else if (isFreeCall(U, TLI)) { Writers.push_back(cast(U)->getParent()->getParent()); } else if (CallInst *CI = dyn_cast(U)) { // Make sure that this is just the function being called, not that it is @@ -329,7 +329,7 @@ bool GlobalsModRef::AnalyzeIndirectGlobalMemory(GlobalValue *GV) { // Check the value being stored. Value *Ptr = GetUnderlyingObject(SI->getOperand(0)); - if (!isAllocLikeFn(Ptr)) + if (!isAllocLikeFn(Ptr, TLI)) return false; // Too hard to analyze. // Analyze all uses of the allocation. If any of them are used in a @@ -458,7 +458,7 @@ void GlobalsModRef::AnalyzeCallGraph(CallGraph &CG, Module &M) { if (SI->isVolatile()) // Treat volatile stores as reading memory somewhere. FunctionEffect |= Ref; - } else if (isAllocationFn(&*II) || isFreeCall(&*II)) { + } else if (isAllocationFn(&*II, TLI) || isFreeCall(&*II, TLI)) { FunctionEffect |= ModRef; } else if (IntrinsicInst *Intrinsic = dyn_cast(&*II)) { // The callgraph doesn't include intrinsic calls. diff --git a/lib/Analysis/MemoryBuiltins.cpp b/lib/Analysis/MemoryBuiltins.cpp index b986b32..5b2313e 100644 --- a/lib/Analysis/MemoryBuiltins.cpp +++ b/lib/Analysis/MemoryBuiltins.cpp @@ -26,6 +26,7 @@ #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetLibraryInfo.h" #include "llvm/Transforms/Utils/Local.h" using namespace llvm; @@ -39,7 +40,7 @@ enum AllocType { }; struct AllocFnsTy { - const char *Name; + LibFunc::Func Func; AllocType AllocTy; unsigned char NumParams; // First and Second size parameters (or -1 if unused) @@ -49,22 +50,22 @@ struct AllocFnsTy { // FIXME: certain users need more information. E.g., SimplifyLibCalls needs to // know which functions are nounwind, noalias, nocapture parameters, etc. static const AllocFnsTy AllocationFnData[] = { - {"malloc", MallocLike, 1, 0, -1}, - {"valloc", MallocLike, 1, 0, -1}, - {"_Znwj", MallocLike, 1, 0, -1}, // new(unsigned int) - {"_ZnwjRKSt9nothrow_t", MallocLike, 2, 0, -1}, // new(unsigned int, nothrow) - {"_Znwm", MallocLike, 1, 0, -1}, // new(unsigned long) - {"_ZnwmRKSt9nothrow_t", MallocLike, 2, 0, -1}, // new(unsigned long, nothrow) - {"_Znaj", MallocLike, 1, 0, -1}, // new[](unsigned int) - {"_ZnajRKSt9nothrow_t", MallocLike, 2, 0, -1}, // new[](unsigned int, nothrow) - {"_Znam", MallocLike, 1, 0, -1}, // new[](unsigned long) - {"_ZnamRKSt9nothrow_t", MallocLike, 2, 0, -1}, // new[](unsigned long, nothrow) - {"posix_memalign", MallocLike, 3, 2, -1}, - {"calloc", CallocLike, 2, 0, 1}, - {"realloc", ReallocLike, 2, 1, -1}, - {"reallocf", ReallocLike, 2, 1, -1}, - {"strdup", StrDupLike, 1, -1, -1}, - {"strndup", StrDupLike, 2, 1, -1} + {LibFunc::malloc, MallocLike, 1, 0, -1}, + {LibFunc::valloc, MallocLike, 1, 0, -1}, + {LibFunc::Znwj, MallocLike, 1, 0, -1}, // new(unsigned int) + {LibFunc::ZnwjRKSt9nothrow_t, MallocLike, 2, 0, -1}, // new(unsigned int, nothrow) + {LibFunc::Znwm, MallocLike, 1, 0, -1}, // new(unsigned long) + {LibFunc::ZnwmRKSt9nothrow_t, MallocLike, 2, 0, -1}, // new(unsigned long, nothrow) + {LibFunc::Znaj, MallocLike, 1, 0, -1}, // new[](unsigned int) + {LibFunc::ZnajRKSt9nothrow_t, MallocLike, 2, 0, -1}, // new[](unsigned int, nothrow) + {LibFunc::Znam, MallocLike, 1, 0, -1}, // new[](unsigned long) + {LibFunc::ZnamRKSt9nothrow_t, MallocLike, 2, 0, -1}, // new[](unsigned long, nothrow) + {LibFunc::posix_memalign, MallocLike, 3, 2, -1}, + {LibFunc::calloc, CallocLike, 2, 0, 1}, + {LibFunc::realloc, ReallocLike, 2, 1, -1}, + {LibFunc::reallocf, ReallocLike, 2, 1, -1}, + {LibFunc::strdup, StrDupLike, 1, -1, -1}, + {LibFunc::strndup, StrDupLike, 2, 1, -1} }; @@ -85,15 +86,22 @@ static Function *getCalledFunction(const Value *V, bool LookThroughBitCast) { /// \brief Returns the allocation data for the given value if it is a call to a /// known allocation function, and NULL otherwise. static const AllocFnsTy *getAllocationData(const Value *V, AllocType AllocTy, + const TargetLibraryInfo *TLI, bool LookThroughBitCast = false) { Function *Callee = getCalledFunction(V, LookThroughBitCast); if (!Callee) return 0; + // Make sure that the function is available. + StringRef FnName = Callee->getName(); + LibFunc::Func TLIFn; + if (!TLI || !TLI->getLibFunc(FnName, TLIFn) || !TLI->has(TLIFn)) + return 0; + unsigned i = 0; bool found = false; for ( ; i < array_lengthof(AllocationFnData); ++i) { - if (Callee->getName() == AllocationFnData[i].Name) { + if (AllocationFnData[i].Func == TLIFn) { found = true; break; } @@ -106,7 +114,6 @@ static const AllocFnsTy *getAllocationData(const Value *V, AllocType AllocTy, return 0; // Check function prototype. - // FIXME: Check the nobuiltin metadata?? (PR5130) int FstParam = FnData->FstParam; int SndParam = FnData->SndParam; FunctionType *FTy = Callee->getFunctionType(); @@ -132,57 +139,65 @@ static bool hasNoAliasAttr(const Value *V, bool LookThroughBitCast) { /// \brief Tests if a value is a call or invoke to a library function that /// allocates or reallocates memory (either malloc, calloc, realloc, or strdup /// like). -bool llvm::isAllocationFn(const Value *V, bool LookThroughBitCast) { - return getAllocationData(V, AnyAlloc, LookThroughBitCast); +bool llvm::isAllocationFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast) { + return getAllocationData(V, AnyAlloc, TLI, LookThroughBitCast); } /// \brief Tests if a value is a call or invoke to a function that returns a /// NoAlias pointer (including malloc/calloc/realloc/strdup-like functions). -bool llvm::isNoAliasFn(const Value *V, bool LookThroughBitCast) { +bool llvm::isNoAliasFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast) { // it's safe to consider realloc as noalias since accessing the original // pointer is undefined behavior - return isAllocationFn(V, LookThroughBitCast) || + return isAllocationFn(V, TLI, LookThroughBitCast) || hasNoAliasAttr(V, LookThroughBitCast); } /// \brief Tests if a value is a call or invoke to a library function that /// allocates uninitialized memory (such as malloc). -bool llvm::isMallocLikeFn(const Value *V, bool LookThroughBitCast) { - return getAllocationData(V, MallocLike, LookThroughBitCast); +bool llvm::isMallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast) { + return getAllocationData(V, MallocLike, TLI, LookThroughBitCast); } /// \brief Tests if a value is a call or invoke to a library function that /// allocates zero-filled memory (such as calloc). -bool llvm::isCallocLikeFn(const Value *V, bool LookThroughBitCast) { - return getAllocationData(V, CallocLike, LookThroughBitCast); +bool llvm::isCallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast) { + return getAllocationData(V, CallocLike, TLI, LookThroughBitCast); } /// \brief Tests if a value is a call or invoke to a library function that /// allocates memory (either malloc, calloc, or strdup like). -bool llvm::isAllocLikeFn(const Value *V, bool LookThroughBitCast) { - return getAllocationData(V, AllocLike, LookThroughBitCast); +bool llvm::isAllocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast) { + return getAllocationData(V, AllocLike, TLI, LookThroughBitCast); } /// \brief Tests if a value is a call or invoke to a library function that /// reallocates memory (such as realloc). -bool llvm::isReallocLikeFn(const Value *V, bool LookThroughBitCast) { - return getAllocationData(V, ReallocLike, LookThroughBitCast); +bool llvm::isReallocLikeFn(const Value *V, const TargetLibraryInfo *TLI, + bool LookThroughBitCast) { + return getAllocationData(V, ReallocLike, TLI, LookThroughBitCast); } /// extractMallocCall - Returns the corresponding CallInst if the instruction /// is a malloc call. Since CallInst::CreateMalloc() only creates calls, we /// ignore InvokeInst here. -const CallInst *llvm::extractMallocCall(const Value *I) { - return isMallocLikeFn(I) ? dyn_cast(I) : 0; +const CallInst *llvm::extractMallocCall(const Value *I, + const TargetLibraryInfo *TLI) { + return isMallocLikeFn(I, TLI) ? dyn_cast(I) : 0; } static Value *computeArraySize(const CallInst *CI, const TargetData *TD, + const TargetLibraryInfo *TLI, bool LookThroughSExt = false) { if (!CI) return NULL; // The size of the malloc's result type must be known to determine array size. - Type *T = getMallocAllocatedType(CI); + Type *T = getMallocAllocatedType(CI, TLI); if (!T || !T->isSized() || !TD) return NULL; @@ -204,9 +219,11 @@ static Value *computeArraySize(const CallInst *CI, const TargetData *TD, /// isArrayMalloc - Returns the corresponding CallInst if the instruction /// is a call to malloc whose array size can be determined and the array size /// is not constant 1. Otherwise, return NULL. -const CallInst *llvm::isArrayMalloc(const Value *I, const TargetData *TD) { - const CallInst *CI = extractMallocCall(I); - Value *ArraySize = computeArraySize(CI, TD); +const CallInst *llvm::isArrayMalloc(const Value *I, + const TargetData *TD, + const TargetLibraryInfo *TLI) { + const CallInst *CI = extractMallocCall(I, TLI); + Value *ArraySize = computeArraySize(CI, TD, TLI); if (ArraySize && ArraySize != ConstantInt::get(CI->getArgOperand(0)->getType(), 1)) @@ -221,8 +238,9 @@ const CallInst *llvm::isArrayMalloc(const Value *I, const TargetData *TD) { /// 0: PointerType is the calls' return type. /// 1: PointerType is the bitcast's result type. /// >1: Unique PointerType cannot be determined, return NULL. -PointerType *llvm::getMallocType(const CallInst *CI) { - assert(isMallocLikeFn(CI) && "getMallocType and not malloc call"); +PointerType *llvm::getMallocType(const CallInst *CI, + const TargetLibraryInfo *TLI) { + assert(isMallocLikeFn(CI, TLI) && "getMallocType and not malloc call"); PointerType *MallocType = NULL; unsigned NumOfBitCastUses = 0; @@ -252,8 +270,9 @@ PointerType *llvm::getMallocType(const CallInst *CI) { /// 0: PointerType is the malloc calls' return type. /// 1: PointerType is the bitcast's result type. /// >1: Unique PointerType cannot be determined, return NULL. -Type *llvm::getMallocAllocatedType(const CallInst *CI) { - PointerType *PT = getMallocType(CI); +Type *llvm::getMallocAllocatedType(const CallInst *CI, + const TargetLibraryInfo *TLI) { + PointerType *PT = getMallocType(CI, TLI); return PT ? PT->getElementType() : NULL; } @@ -263,21 +282,23 @@ Type *llvm::getMallocAllocatedType(const CallInst *CI) { /// constant 1. Otherwise, return NULL for mallocs whose array size cannot be /// determined. Value *llvm::getMallocArraySize(CallInst *CI, const TargetData *TD, + const TargetLibraryInfo *TLI, bool LookThroughSExt) { - assert(isMallocLikeFn(CI) && "getMallocArraySize and not malloc call"); - return computeArraySize(CI, TD, LookThroughSExt); + assert(isMallocLikeFn(CI, TLI) && "getMallocArraySize and not malloc call"); + return computeArraySize(CI, TD, TLI, LookThroughSExt); } /// extractCallocCall - Returns the corresponding CallInst if the instruction /// is a calloc call. -const CallInst *llvm::extractCallocCall(const Value *I) { - return isCallocLikeFn(I) ? cast(I) : 0; +const CallInst *llvm::extractCallocCall(const Value *I, + const TargetLibraryInfo *TLI) { + return isCallocLikeFn(I, TLI) ? cast(I) : 0; } /// isFreeCall - Returns non-null if the value is a call to the builtin free() -const CallInst *llvm::isFreeCall(const Value *I) { +const CallInst *llvm::isFreeCall(const Value *I, const TargetLibraryInfo *TLI) { const CallInst *CI = dyn_cast(I); if (!CI) return 0; @@ -285,9 +306,14 @@ const CallInst *llvm::isFreeCall(const Value *I) { if (Callee == 0 || !Callee->isDeclaration()) return 0; - if (Callee->getName() != "free" && - Callee->getName() != "_ZdlPv" && // operator delete(void*) - Callee->getName() != "_ZdaPv") // operator delete[](void*) + StringRef FnName = Callee->getName(); + LibFunc::Func TLIFn; + if (!TLI || !TLI->getLibFunc(FnName, TLIFn) || !TLI->has(TLIFn)) + return 0; + + if (TLIFn != LibFunc::free && + TLIFn != LibFunc::ZdlPv && // operator delete(void*) + TLIFn != LibFunc::ZdaPv) // operator delete[](void*) return 0; // Check free prototype. @@ -316,11 +342,11 @@ const CallInst *llvm::isFreeCall(const Value *I) { /// If RoundToAlign is true, then Size is rounded up to the aligment of allocas, /// byval arguments, and global variables. bool llvm::getObjectSize(const Value *Ptr, uint64_t &Size, const TargetData *TD, - bool RoundToAlign) { + const TargetLibraryInfo *TLI, bool RoundToAlign) { if (!TD) return false; - ObjectSizeOffsetVisitor Visitor(TD, Ptr->getContext(), RoundToAlign); + ObjectSizeOffsetVisitor Visitor(TD, TLI, Ptr->getContext(), RoundToAlign); SizeOffsetType Data = Visitor.compute(const_cast(Ptr)); if (!Visitor.bothKnown(Data)) return false; @@ -348,9 +374,10 @@ APInt ObjectSizeOffsetVisitor::align(APInt Size, uint64_t Align) { } ObjectSizeOffsetVisitor::ObjectSizeOffsetVisitor(const TargetData *TD, + const TargetLibraryInfo *TLI, LLVMContext &Context, bool RoundToAlign) -: TD(TD), RoundToAlign(RoundToAlign) { +: TD(TD), TLI(TLI), RoundToAlign(RoundToAlign) { IntegerType *IntTy = TD->getIntPtrType(Context); IntTyBits = IntTy->getBitWidth(); Zero = APInt::getNullValue(IntTyBits); @@ -416,7 +443,8 @@ SizeOffsetType ObjectSizeOffsetVisitor::visitArgument(Argument &A) { } SizeOffsetType ObjectSizeOffsetVisitor::visitCallSite(CallSite CS) { - const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc); + const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc, + TLI); if (!FnData) return unknown(); @@ -532,8 +560,9 @@ SizeOffsetType ObjectSizeOffsetVisitor::visitInstruction(Instruction &I) { ObjectSizeOffsetEvaluator::ObjectSizeOffsetEvaluator(const TargetData *TD, + const TargetLibraryInfo *TLI, LLVMContext &Context) -: TD(TD), Context(Context), Builder(Context, TargetFolder(TD)) { +: TD(TD), TLI(TLI), Context(Context), Builder(Context, TargetFolder(TD)) { IntTy = TD->getIntPtrType(Context); Zero = ConstantInt::get(IntTy, 0); } @@ -558,7 +587,7 @@ SizeOffsetEvalType ObjectSizeOffsetEvaluator::compute(Value *V) { } SizeOffsetEvalType ObjectSizeOffsetEvaluator::compute_(Value *V) { - ObjectSizeOffsetVisitor Visitor(TD, Context); + ObjectSizeOffsetVisitor Visitor(TD, TLI, Context); SizeOffsetType Const = Visitor.compute(V); if (Visitor.bothKnown(Const)) return std::make_pair(ConstantInt::get(Context, Const.first), @@ -621,7 +650,8 @@ SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitAllocaInst(AllocaInst &I) { } SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitCallSite(CallSite CS) { - const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc); + const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc, + TLI); if (!FnData) return unknown(); diff --git a/lib/Analysis/MemoryDependenceAnalysis.cpp b/lib/Analysis/MemoryDependenceAnalysis.cpp index 059e574..2accaef 100644 --- a/lib/Analysis/MemoryDependenceAnalysis.cpp +++ b/lib/Analysis/MemoryDependenceAnalysis.cpp @@ -148,7 +148,7 @@ AliasAnalysis::ModRefResult GetLocation(const Instruction *Inst, return AliasAnalysis::ModRef; } - if (const CallInst *CI = isFreeCall(Inst)) { + if (const CallInst *CI = isFreeCall(Inst, AA->getTargetLibraryInfo())) { // calls to free() deallocate the entire structure Loc = AliasAnalysis::Location(CI->getArgOperand(0)); return AliasAnalysis::Mod; @@ -479,7 +479,7 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad, // a subsequent bitcast of the malloc call result. There can be stores to // the malloced memory between the malloc call and its bitcast uses, and we // need to continue scanning until the malloc call. - if (isa(Inst) || isNoAliasFn(Inst)) { + if (isa(Inst) || isNoAliasFn(Inst, AA->getTargetLibraryInfo())){ const Value *AccessPtr = GetUnderlyingObject(MemLoc.Ptr, TD); if (AccessPtr == Inst || AA->isMustAlias(Inst, AccessPtr)) diff --git a/lib/Target/TargetLibraryInfo.cpp b/lib/Target/TargetLibraryInfo.cpp index 924b1ed..62f973e 100644 --- a/lib/Target/TargetLibraryInfo.cpp +++ b/lib/Target/TargetLibraryInfo.cpp @@ -24,6 +24,16 @@ void TargetLibraryInfo::anchor() { } const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] = { + "_ZdaPv", + "_ZdlPv", + "_Znaj", + "_ZnajRKSt9nothrow_t", + "_Znam", + "_ZnamRKSt9nothrow_t", + "_Znwj", + "_ZnwjRKSt9nothrow_t", + "_Znwm", + "_ZnwmRKSt9nothrow_t", "__cxa_atexit", "__cxa_guard_abort", "__cxa_guard_acquire", @@ -50,6 +60,7 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] = "atanhf", "atanhl", "atanl", + "calloc", "cbrt", "cbrtf", "cbrtl", @@ -89,6 +100,7 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] = "fmodl", "fputc", "fputs", + "free", "fwrite", "iprintf", "log", @@ -106,6 +118,7 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] = "logbl", "logf", "logl", + "malloc", "memchr", "memcmp", "memcpy", @@ -115,11 +128,14 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] = "nearbyint", "nearbyintf", "nearbyintl", + "posix_memalign", "pow", "powf", "powl", "putchar", "puts", + "realloc", + "reallocf", "rint", "rintf", "rintl", @@ -139,10 +155,12 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] = "strcat", "strchr", "strcpy", + "strdup", "strlen", "strncat", "strncmp", "strncpy", + "strndup", "strnlen", "tan", "tanf", @@ -152,7 +170,8 @@ const char* TargetLibraryInfo::StandardNames[LibFunc::NumLibFuncs] = "tanl", "trunc", "truncf", - "truncl" + "truncl", + "valloc" }; /// initialize - Initialize the set of available library functions based on the diff --git a/lib/Transforms/IPO/GlobalOpt.cpp b/lib/Transforms/IPO/GlobalOpt.cpp index 6d950d2..b888e95 100644 --- a/lib/Transforms/IPO/GlobalOpt.cpp +++ b/lib/Transforms/IPO/GlobalOpt.cpp @@ -346,7 +346,7 @@ static bool isLeakCheckerRoot(GlobalVariable *GV) { /// Given a value that is stored to a global but never read, determine whether /// it's safe to remove the store and the chain of computation that feeds the /// store. -static bool IsSafeComputationToRemove(Value *V) { +static bool IsSafeComputationToRemove(Value *V, const TargetLibraryInfo *TLI) { do { if (isa(V)) return true; @@ -355,7 +355,7 @@ static bool IsSafeComputationToRemove(Value *V) { if (isa(V) || isa(V) || isa(V) || isa(V)) return false; - if (isAllocationFn(V)) + if (isAllocationFn(V, TLI)) return true; Instruction *I = cast(V); @@ -376,7 +376,8 @@ static bool IsSafeComputationToRemove(Value *V) { /// of the global and clean up any that obviously don't assign the global a /// value that isn't dynamically allocated. /// -static bool CleanupPointerRootUsers(GlobalVariable *GV) { +static bool CleanupPointerRootUsers(GlobalVariable *GV, + const TargetLibraryInfo *TLI) { // A brief explanation of leak checkers. The goal is to find bugs where // pointers are forgotten, causing an accumulating growth in memory // usage over time. The common strategy for leak checkers is to whitelist the @@ -432,18 +433,18 @@ static bool CleanupPointerRootUsers(GlobalVariable *GV) { C->destroyConstant(); // This could have invalidated UI, start over from scratch. Dead.clear(); - CleanupPointerRootUsers(GV); + CleanupPointerRootUsers(GV, TLI); return true; } } } for (int i = 0, e = Dead.size(); i != e; ++i) { - if (IsSafeComputationToRemove(Dead[i].first)) { + if (IsSafeComputationToRemove(Dead[i].first, TLI)) { Dead[i].second->eraseFromParent(); Instruction *I = Dead[i].first; do { - if (isAllocationFn(I)) + if (isAllocationFn(I, TLI)) break; Instruction *J = dyn_cast(I->getOperand(0)); if (!J) @@ -975,7 +976,7 @@ static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV, // nor is the global. if (AllNonStoreUsesGone) { if (isLeakCheckerRoot(GV)) { - Changed |= CleanupPointerRootUsers(GV); + Changed |= CleanupPointerRootUsers(GV, TLI); } else { Changed = true; CleanupConstantGlobalUsers(GV, 0, TD, TLI); @@ -1465,9 +1466,10 @@ static void RewriteUsesOfLoadForHeapSRoA(LoadInst *Load, /// PerformHeapAllocSRoA - CI is an allocation of an array of structures. Break /// it up into multiple allocations of arrays of the fields. static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, - Value *NElems, TargetData *TD) { + Value *NElems, TargetData *TD, + const TargetLibraryInfo *TLI) { DEBUG(dbgs() << "SROA HEAP ALLOC: " << *GV << " MALLOC = " << *CI << '\n'); - Type *MAT = getMallocAllocatedType(CI); + Type *MAT = getMallocAllocatedType(CI, TLI); StructType *STy = cast(MAT); // There is guaranteed to be at least one use of the malloc (storing @@ -1688,7 +1690,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, // This eliminates dynamic allocation, avoids an indirection accessing the // data, and exposes the resultant global to further GlobalOpt. // We cannot optimize the malloc if we cannot determine malloc array size. - Value *NElems = getMallocArraySize(CI, TD, true); + Value *NElems = getMallocArraySize(CI, TD, TLI, true); if (!NElems) return false; @@ -1725,7 +1727,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, // If this is a fixed size array, transform the Malloc to be an alloc of // structs. malloc [100 x struct],1 -> malloc struct, 100 - if (ArrayType *AT = dyn_cast(getMallocAllocatedType(CI))) { + if (ArrayType *AT = dyn_cast(getMallocAllocatedType(CI, TLI))) { Type *IntPtrTy = TD->getIntPtrType(CI->getContext()); unsigned TypeSize = TD->getStructLayout(AllocSTy)->getSizeInBytes(); Value *AllocSize = ConstantInt::get(IntPtrTy, TypeSize); @@ -1742,7 +1744,8 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, CI = cast(Malloc); } - GVI = PerformHeapAllocSRoA(GV, CI, getMallocArraySize(CI, TD, true), TD); + GVI = PerformHeapAllocSRoA(GV, CI, getMallocArraySize(CI, TD, TLI, true), + TD, TLI); return true; } @@ -1771,8 +1774,8 @@ static bool OptimizeOnceStoredGlobal(GlobalVariable *GV, Value *StoredOnceVal, // Optimize away any trapping uses of the loaded value. if (OptimizeAwayTrappingUsesOfLoads(GV, SOVC, TD, TLI)) return true; - } else if (CallInst *CI = extractMallocCall(StoredOnceVal)) { - Type *MallocType = getMallocAllocatedType(CI); + } else if (CallInst *CI = extractMallocCall(StoredOnceVal, TLI)) { + Type *MallocType = getMallocAllocatedType(CI, TLI); if (MallocType && TryToOptimizeStoreOfMallocToGlobal(GV, CI, MallocType, Ordering, GVI, TD, TLI)) @@ -1964,7 +1967,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, bool Changed; if (isLeakCheckerRoot(GV)) { // Delete any constant stores to the global. - Changed = CleanupPointerRootUsers(GV); + Changed = CleanupPointerRootUsers(GV, TLI); } else { // Delete any stores we can find to the global. We may not be able to // make it completely dead though. diff --git a/lib/Transforms/IPO/Inliner.cpp b/lib/Transforms/IPO/Inliner.cpp index 712888a..69a22fb 100644 --- a/lib/Transforms/IPO/Inliner.cpp +++ b/lib/Transforms/IPO/Inliner.cpp @@ -20,6 +20,7 @@ #include "llvm/Analysis/CallGraph.h" #include "llvm/Analysis/InlineCost.h" #include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetLibraryInfo.h" #include "llvm/Transforms/IPO/InlinerPass.h" #include "llvm/Transforms/Utils/Cloning.h" #include "llvm/Transforms/Utils/Local.h" @@ -339,6 +340,7 @@ static bool InlineHistoryIncludes(Function *F, int InlineHistoryID, bool Inliner::runOnSCC(CallGraphSCC &SCC) { CallGraph &CG = getAnalysis(); const TargetData *TD = getAnalysisIfAvailable(); + const TargetLibraryInfo *TLI = getAnalysisIfAvailable(); SmallPtrSet SCCFunctions; DEBUG(dbgs() << "Inliner visiting SCC:"); @@ -417,7 +419,7 @@ bool Inliner::runOnSCC(CallGraphSCC &SCC) { // just delete the call instead of trying to inline it, regardless of // size. This happens because IPSCCP propagates the result out of the // call and then we're left with the dead call. - if (isInstructionTriviallyDead(CS.getInstruction())) { + if (isInstructionTriviallyDead(CS.getInstruction(), TLI)) { DEBUG(dbgs() << " -> Deleting dead call: " << *CS.getInstruction() << "\n"); // Update the call graph by deleting the edge from Callee to Caller. diff --git a/lib/Transforms/InstCombine/InstCombineCalls.cpp b/lib/Transforms/InstCombine/InstCombineCalls.cpp index cbe1ca4..b12fc01 100644 --- a/lib/Transforms/InstCombine/InstCombineCalls.cpp +++ b/lib/Transforms/InstCombine/InstCombineCalls.cpp @@ -168,7 +168,7 @@ Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) { /// the heavy lifting. /// Instruction *InstCombiner::visitCallInst(CallInst &CI) { - if (isFreeCall(&CI)) + if (isFreeCall(&CI, TLI)) return visitFree(CI); // If the caller function is nounwind, mark the call as nounwind, even if the @@ -243,7 +243,7 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) { default: break; case Intrinsic::objectsize: { uint64_t Size; - if (getObjectSize(II->getArgOperand(0), Size, TD)) + if (getObjectSize(II->getArgOperand(0), Size, TD, TLI)) return ReplaceInstUsesWith(CI, ConstantInt::get(CI.getType(), Size)); return 0; } @@ -877,7 +877,7 @@ static IntrinsicInst *FindInitTrampoline(Value *Callee) { // visitCallSite - Improvements for call and invoke instructions. // Instruction *InstCombiner::visitCallSite(CallSite CS) { - if (isAllocLikeFn(CS.getInstruction())) + if (isAllocLikeFn(CS.getInstruction(), TLI)) return visitAllocSite(*CS.getInstruction()); bool Changed = false; diff --git a/lib/Transforms/InstCombine/InstructionCombining.cpp b/lib/Transforms/InstCombine/InstructionCombining.cpp index 68ecd51..ff758c4 100644 --- a/lib/Transforms/InstCombine/InstructionCombining.cpp +++ b/lib/Transforms/InstCombine/InstructionCombining.cpp @@ -1068,7 +1068,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) { // If the bitcast is of an allocation, and the allocation will be // converted to match the type of the cast, don't touch this. if (isa(BCI->getOperand(0)) || - isAllocationFn(BCI->getOperand(0))) { + isAllocationFn(BCI->getOperand(0), TLI)) { // See if the bitcast simplifies, if so, don't nuke this GEP yet. if (Instruction *I = visitBitCast(*BCI)) { if (I != BCI) { @@ -1107,7 +1107,8 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) { static bool -isAllocSiteRemovable(Instruction *AI, SmallVectorImpl &Users) { +isAllocSiteRemovable(Instruction *AI, SmallVectorImpl &Users, + const TargetLibraryInfo *TLI) { SmallVector Worklist; Worklist.push_back(AI); @@ -1163,7 +1164,7 @@ isAllocSiteRemovable(Instruction *AI, SmallVectorImpl &Users) { } } - if (isFreeCall(I)) { + if (isFreeCall(I, TLI)) { Users.push_back(I); continue; } @@ -1188,7 +1189,7 @@ Instruction *InstCombiner::visitAllocSite(Instruction &MI) { // to null and free calls, delete the calls and replace the comparisons with // true or false as appropriate. SmallVector Users; - if (isAllocSiteRemovable(&MI, Users)) { + if (isAllocSiteRemovable(&MI, Users, TLI)) { for (unsigned i = 0, e = Users.size(); i != e; ++i) { Instruction *I = cast_or_null(&*Users[i]); if (!I) continue; @@ -1872,7 +1873,7 @@ static bool AddReachableCodeToWorklist(BasicBlock *BB, Instruction *Inst = BBI++; // DCE instruction if trivially dead. - if (isInstructionTriviallyDead(Inst)) { + if (isInstructionTriviallyDead(Inst, TLI)) { ++NumDeadInst; DEBUG(errs() << "IC: DCE: " << *Inst << '\n'); Inst->eraseFromParent(); @@ -2002,7 +2003,7 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) { if (I == 0) continue; // skip null values. // Check to see if we can DCE the instruction. - if (isInstructionTriviallyDead(I)) { + if (isInstructionTriviallyDead(I, TLI)) { DEBUG(errs() << "IC: DCE: " << *I << '\n'); EraseInstFromFunction(*I); ++NumDeadInst; @@ -2102,7 +2103,7 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) { // If the instruction was modified, it's possible that it is now dead. // if so, remove it. - if (isInstructionTriviallyDead(I)) { + if (isInstructionTriviallyDead(I, TLI)) { EraseInstFromFunction(*I); } else { Worklist.Add(I); diff --git a/lib/Transforms/Instrumentation/BoundsChecking.cpp b/lib/Transforms/Instrumentation/BoundsChecking.cpp index 09e0f14..6429081 100644 --- a/lib/Transforms/Instrumentation/BoundsChecking.cpp +++ b/lib/Transforms/Instrumentation/BoundsChecking.cpp @@ -24,6 +24,7 @@ #include "llvm/Support/TargetFolder.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetLibraryInfo.h" #include "llvm/Transforms/Instrumentation.h" using namespace llvm; @@ -48,10 +49,12 @@ namespace { virtual void getAnalysisUsage(AnalysisUsage &AU) const { AU.addRequired(); + AU.addRequired(); } private: const TargetData *TD; + const TargetLibraryInfo *TLI; ObjectSizeOffsetEvaluator *ObjSizeEval; BuilderTy *Builder; Instruction *Inst; @@ -166,11 +169,12 @@ bool BoundsChecking::instrument(Value *Ptr, Value *InstVal) { bool BoundsChecking::runOnFunction(Function &F) { TD = &getAnalysis(); + TLI = &getAnalysis(); TrapBB = 0; BuilderTy TheBuilder(F.getContext(), TargetFolder(TD)); Builder = &TheBuilder; - ObjectSizeOffsetEvaluator TheObjSizeEval(TD, F.getContext()); + ObjectSizeOffsetEvaluator TheObjSizeEval(TD, TLI, F.getContext()); ObjSizeEval = &TheObjSizeEval; // check HANDLE_MEMORY_INST in include/llvm/Instruction.def for memory diff --git a/lib/Transforms/Scalar/CodeGenPrepare.cpp b/lib/Transforms/Scalar/CodeGenPrepare.cpp index a8deda8..cfca726 100644 --- a/lib/Transforms/Scalar/CodeGenPrepare.cpp +++ b/lib/Transforms/Scalar/CodeGenPrepare.cpp @@ -988,7 +988,7 @@ bool CodeGenPrepare::OptimizeMemoryInst(Instruction *MemoryInst, Value *Addr, WeakVH IterHandle(CurInstIterator); BasicBlock *BB = CurInstIterator->getParent(); - RecursivelyDeleteTriviallyDeadInstructions(Repl); + RecursivelyDeleteTriviallyDeadInstructions(Repl, TLInfo); if (IterHandle != CurInstIterator) { // If the iterator instruction was recursively deleted, start over at the diff --git a/lib/Transforms/Scalar/DCE.cpp b/lib/Transforms/Scalar/DCE.cpp index 8dbcc23..086f0a1 100644 --- a/lib/Transforms/Scalar/DCE.cpp +++ b/lib/Transforms/Scalar/DCE.cpp @@ -22,6 +22,7 @@ #include "llvm/Instruction.h" #include "llvm/Pass.h" #include "llvm/Support/InstIterator.h" +#include "llvm/Target/TargetLibraryInfo.h" #include "llvm/ADT/Statistic.h" using namespace llvm; @@ -38,10 +39,11 @@ namespace { initializeDeadInstEliminationPass(*PassRegistry::getPassRegistry()); } virtual bool runOnBasicBlock(BasicBlock &BB) { + TargetLibraryInfo *TLI = getAnalysisIfAvailable(); bool Changed = false; for (BasicBlock::iterator DI = BB.begin(); DI != BB.end(); ) { Instruction *Inst = DI++; - if (isInstructionTriviallyDead(Inst)) { + if (isInstructionTriviallyDead(Inst, TLI)) { Inst->eraseFromParent(); Changed = true; ++DIEEliminated; @@ -87,6 +89,8 @@ char DCE::ID = 0; INITIALIZE_PASS(DCE, "dce", "Dead Code Elimination", false, false) bool DCE::runOnFunction(Function &F) { + TargetLibraryInfo *TLI = getAnalysisIfAvailable(); + // Start out with all of the instructions in the worklist... std::vector WorkList; for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i) @@ -101,7 +105,7 @@ bool DCE::runOnFunction(Function &F) { Instruction *I = WorkList.back(); WorkList.pop_back(); - if (isInstructionTriviallyDead(I)) { // If the instruction is dead. + if (isInstructionTriviallyDead(I, TLI)) { // If the instruction is dead. // Loop over all of the values that the instruction uses, if there are // instructions being used, add them to the worklist, because they might // go dead after this one is removed. diff --git a/lib/Transforms/Scalar/DeadStoreElimination.cpp b/lib/Transforms/Scalar/DeadStoreElimination.cpp index 8b1283f..25a1dd7 100644 --- a/lib/Transforms/Scalar/DeadStoreElimination.cpp +++ b/lib/Transforms/Scalar/DeadStoreElimination.cpp @@ -106,6 +106,7 @@ FunctionPass *llvm::createDeadStoreEliminationPass() { return new DSE(); } /// static void DeleteDeadInstruction(Instruction *I, MemoryDependenceAnalysis &MD, + const TargetLibraryInfo *TLI, SmallSetVector *ValueSet = 0) { SmallVector NowDeadInsts; @@ -130,7 +131,7 @@ static void DeleteDeadInstruction(Instruction *I, if (!Op->use_empty()) continue; if (Instruction *OpI = dyn_cast(Op)) - if (isInstructionTriviallyDead(OpI)) + if (isInstructionTriviallyDead(OpI, TLI)) NowDeadInsts.push_back(OpI); } @@ -276,7 +277,7 @@ static Value *getStoredPointerOperand(Instruction *I) { static uint64_t getPointerSize(const Value *V, AliasAnalysis &AA) { uint64_t Size; - if (getObjectSize(V, Size, AA.getTargetData())) + if (getObjectSize(V, Size, AA.getTargetData(), AA.getTargetLibraryInfo())) return Size; return AliasAnalysis::UnknownSize; } @@ -454,7 +455,7 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) { Instruction *Inst = BBI++; // Handle 'free' calls specially. - if (CallInst *F = isFreeCall(Inst)) { + if (CallInst *F = isFreeCall(Inst, AA->getTargetLibraryInfo())) { MadeChange |= HandleFree(F); continue; } @@ -483,7 +484,7 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) { // in case we need it. WeakVH NextInst(BBI); - DeleteDeadInstruction(SI, *MD); + DeleteDeadInstruction(SI, *MD, AA->getTargetLibraryInfo()); if (NextInst == 0) // Next instruction deleted. BBI = BB.begin(); @@ -530,7 +531,7 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) { << *DepWrite << "\n KILLER: " << *Inst << '\n'); // Delete the store and now-dead instructions that feed it. - DeleteDeadInstruction(DepWrite, *MD); + DeleteDeadInstruction(DepWrite, *MD, AA->getTargetLibraryInfo()); ++NumFastStores; MadeChange = true; @@ -640,7 +641,7 @@ bool DSE::HandleFree(CallInst *F) { Instruction *Next = llvm::next(BasicBlock::iterator(Dependency)); // DCE instructions only used to calculate that store - DeleteDeadInstruction(Dependency, *MD); + DeleteDeadInstruction(Dependency, *MD, AA->getTargetLibraryInfo()); ++NumFastStores; MadeChange = true; @@ -680,7 +681,8 @@ bool DSE::handleEndBlock(BasicBlock &BB) { // Okay, so these are dead heap objects, but if the pointer never escapes // then it's leaked by this function anyways. - else if (isAllocLikeFn(I) && !PointerMayBeCaptured(I, true, true)) + else if (isAllocLikeFn(I, AA->getTargetLibraryInfo()) && + !PointerMayBeCaptured(I, true, true)) DeadStackObjects.insert(I); } @@ -724,7 +726,8 @@ bool DSE::handleEndBlock(BasicBlock &BB) { dbgs() << '\n'); // DCE instructions only used to calculate that store. - DeleteDeadInstruction(Dead, *MD, &DeadStackObjects); + DeleteDeadInstruction(Dead, *MD, AA->getTargetLibraryInfo(), + &DeadStackObjects); ++NumFastStores; MadeChange = true; continue; @@ -732,9 +735,10 @@ bool DSE::handleEndBlock(BasicBlock &BB) { } // Remove any dead non-memory-mutating instructions. - if (isInstructionTriviallyDead(BBI)) { + if (isInstructionTriviallyDead(BBI, AA->getTargetLibraryInfo())) { Instruction *Inst = BBI++; - DeleteDeadInstruction(Inst, *MD, &DeadStackObjects); + DeleteDeadInstruction(Inst, *MD, AA->getTargetLibraryInfo(), + &DeadStackObjects); ++NumFastOther; MadeChange = true; continue; @@ -750,7 +754,7 @@ bool DSE::handleEndBlock(BasicBlock &BB) { if (CallSite CS = cast(BBI)) { // Remove allocation function calls from the list of dead stack objects; // there can't be any references before the definition. - if (isAllocLikeFn(BBI)) + if (isAllocLikeFn(BBI, AA->getTargetLibraryInfo())) DeadStackObjects.remove(BBI); // If this call does not access memory, it can't be loading any of our diff --git a/lib/Transforms/Scalar/EarlyCSE.cpp b/lib/Transforms/Scalar/EarlyCSE.cpp index 9759549..2627113 100644 --- a/lib/Transforms/Scalar/EarlyCSE.cpp +++ b/lib/Transforms/Scalar/EarlyCSE.cpp @@ -374,7 +374,7 @@ bool EarlyCSE::processNode(DomTreeNode *Node) { Instruction *Inst = I++; // Dead instructions should just be removed. - if (isInstructionTriviallyDead(Inst)) { + if (isInstructionTriviallyDead(Inst, TLI)) { DEBUG(dbgs() << "EarlyCSE DCE: " << *Inst << '\n'); Inst->eraseFromParent(); Changed = true; diff --git a/lib/Transforms/Scalar/GVN.cpp b/lib/Transforms/Scalar/GVN.cpp index 3a89623..558bd35 100644 --- a/lib/Transforms/Scalar/GVN.cpp +++ b/lib/Transforms/Scalar/GVN.cpp @@ -1436,7 +1436,7 @@ bool GVN::processNonLocalLoad(LoadInst *LI) { Instruction *DepInst = DepInfo.getInst(); // Loading the allocation -> undef. - if (isa(DepInst) || isMallocLikeFn(DepInst) || + if (isa(DepInst) || isMallocLikeFn(DepInst, TLI) || // Loading immediately after lifetime begin -> undef. isLifetimeStart(DepInst)) { ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, @@ -1951,7 +1951,7 @@ bool GVN::processLoad(LoadInst *L) { // If this load really doesn't depend on anything, then we must be loading an // undef value. This can happen when loading for a fresh allocation with no // intervening stores, for example. - if (isa(DepInst) || isMallocLikeFn(DepInst)) { + if (isa(DepInst) || isMallocLikeFn(DepInst, TLI)) { L->replaceAllUsesWith(UndefValue::get(L->getType())); markInstructionForDeletion(L); ++NumGVNLoad; diff --git a/lib/Transforms/Scalar/IndVarSimplify.cpp b/lib/Transforms/Scalar/IndVarSimplify.cpp index 37f8bdf..c933a17 100644 --- a/lib/Transforms/Scalar/IndVarSimplify.cpp +++ b/lib/Transforms/Scalar/IndVarSimplify.cpp @@ -44,6 +44,7 @@ #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/SimplifyIndVar.h" #include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetLibraryInfo.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/Statistic.h" @@ -68,6 +69,7 @@ namespace { ScalarEvolution *SE; DominatorTree *DT; TargetData *TD; + TargetLibraryInfo *TLI; SmallVector DeadInsts; bool Changed; @@ -414,11 +416,11 @@ void IndVarSimplify::HandleFloatingPointIV(Loop *L, PHINode *PN) { // new comparison. NewCompare->takeName(Compare); Compare->replaceAllUsesWith(NewCompare); - RecursivelyDeleteTriviallyDeadInstructions(Compare); + RecursivelyDeleteTriviallyDeadInstructions(Compare, TLI); // Delete the old floating point increment. Incr->replaceAllUsesWith(UndefValue::get(Incr->getType())); - RecursivelyDeleteTriviallyDeadInstructions(Incr); + RecursivelyDeleteTriviallyDeadInstructions(Incr, TLI); // If the FP induction variable still has uses, this is because something else // in the loop uses its value. In order to canonicalize the induction @@ -431,7 +433,7 @@ void IndVarSimplify::HandleFloatingPointIV(Loop *L, PHINode *PN) { Value *Conv = new SIToFPInst(NewPHI, PN->getType(), "indvar.conv", PN->getParent()->getFirstInsertionPt()); PN->replaceAllUsesWith(Conv); - RecursivelyDeleteTriviallyDeadInstructions(PN); + RecursivelyDeleteTriviallyDeadInstructions(PN, TLI); } Changed = true; } @@ -550,14 +552,14 @@ void IndVarSimplify::RewriteLoopExitValues(Loop *L, SCEVExpander &Rewriter) { PN->setIncomingValue(i, ExitVal); // If this instruction is dead now, delete it. - RecursivelyDeleteTriviallyDeadInstructions(Inst); + RecursivelyDeleteTriviallyDeadInstructions(Inst, TLI); if (NumPreds == 1) { // Completely replace a single-pred PHI. This is safe, because the // NewVal won't be variant in the loop, so we don't need an LCSSA phi // node anymore. PN->replaceAllUsesWith(ExitVal); - RecursivelyDeleteTriviallyDeadInstructions(PN); + RecursivelyDeleteTriviallyDeadInstructions(PN, TLI); } } if (NumPreds != 1) { @@ -1697,6 +1699,7 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) { SE = &getAnalysis(); DT = &getAnalysis(); TD = getAnalysisIfAvailable(); + TLI = getAnalysisIfAvailable(); DeadInsts.clear(); Changed = false; @@ -1763,7 +1766,7 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) { while (!DeadInsts.empty()) if (Instruction *Inst = dyn_cast_or_null(&*DeadInsts.pop_back_val())) - RecursivelyDeleteTriviallyDeadInstructions(Inst); + RecursivelyDeleteTriviallyDeadInstructions(Inst, TLI); // The Rewriter may not be used from this point on. @@ -1772,7 +1775,7 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) { SinkUnusedInvariants(L); // Clean up dead instructions. - Changed |= DeleteDeadPHIs(L->getHeader()); + Changed |= DeleteDeadPHIs(L->getHeader(), TLI); // Check a post-condition. assert(L->isLCSSAForm(*DT) && "Indvars did not leave the loop in lcssa form!"); diff --git a/lib/Transforms/Scalar/JumpThreading.cpp b/lib/Transforms/Scalar/JumpThreading.cpp index dd42c59..20844c6 100644 --- a/lib/Transforms/Scalar/JumpThreading.cpp +++ b/lib/Transforms/Scalar/JumpThreading.cpp @@ -1455,7 +1455,7 @@ bool JumpThreading::ThreadEdge(BasicBlock *BB, // At this point, the IR is fully up to date and consistent. Do a quick scan // over the new instructions and zap any that are constants or dead. This // frequently happens because of phi translation. - SimplifyInstructionsInBlock(NewBB, TD); + SimplifyInstructionsInBlock(NewBB, TD, TLI); // Threaded an edge! ++NumThreads; diff --git a/lib/Transforms/Scalar/LICM.cpp b/lib/Transforms/Scalar/LICM.cpp index 0192e92..ebff144 100644 --- a/lib/Transforms/Scalar/LICM.cpp +++ b/lib/Transforms/Scalar/LICM.cpp @@ -307,7 +307,7 @@ void LICM::SinkRegion(DomTreeNode *N) { // If the instruction is dead, we would try to sink it because it isn't used // in the loop, instead, just delete it. - if (isInstructionTriviallyDead(&I)) { + if (isInstructionTriviallyDead(&I, TLI)) { DEBUG(dbgs() << "LICM deleting dead inst: " << I << '\n'); ++II; CurAST->deleteValue(&I); diff --git a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp index ac1082c..a72e288 100644 --- a/lib/Transforms/Scalar/LoopIdiomRecognize.cpp +++ b/lib/Transforms/Scalar/LoopIdiomRecognize.cpp @@ -132,7 +132,8 @@ Pass *llvm::createLoopIdiomPass() { return new LoopIdiomRecognize(); } /// and zero out all the operands of this instruction. If any of them become /// dead, delete them and the computation tree that feeds them. /// -static void deleteDeadInstruction(Instruction *I, ScalarEvolution &SE) { +static void deleteDeadInstruction(Instruction *I, ScalarEvolution &SE, + const TargetLibraryInfo *TLI) { SmallVector NowDeadInsts; NowDeadInsts.push_back(I); @@ -153,7 +154,7 @@ static void deleteDeadInstruction(Instruction *I, ScalarEvolution &SE) { if (!Op->use_empty()) continue; if (Instruction *OpI = dyn_cast(Op)) - if (isInstructionTriviallyDead(OpI)) + if (isInstructionTriviallyDead(OpI, TLI)) NowDeadInsts.push_back(OpI); } @@ -164,10 +165,11 @@ static void deleteDeadInstruction(Instruction *I, ScalarEvolution &SE) { /// deleteIfDeadInstruction - If the specified value is a dead instruction, /// delete it and any recursively used instructions. -static void deleteIfDeadInstruction(Value *V, ScalarEvolution &SE) { +static void deleteIfDeadInstruction(Value *V, ScalarEvolution &SE, + const TargetLibraryInfo *TLI) { if (Instruction *I = dyn_cast(V)) - if (isInstructionTriviallyDead(I)) - deleteDeadInstruction(I, SE); + if (isInstructionTriviallyDead(I, TLI)) + deleteDeadInstruction(I, SE, TLI); } bool LoopIdiomRecognize::runOnLoop(Loop *L, LPPassManager &LPM) { @@ -490,7 +492,7 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize, StoreSize, getAnalysis(), TheStore)){ Expander.clear(); // If we generated new code for the base pointer, clean up. - deleteIfDeadInstruction(BasePtr, *SE); + deleteIfDeadInstruction(BasePtr, *SE, TLI); return false; } @@ -538,7 +540,7 @@ processLoopStridedStore(Value *DestPtr, unsigned StoreSize, // Okay, the memset has been formed. Zap the original store and anything that // feeds into it. - deleteDeadInstruction(TheStore, *SE); + deleteDeadInstruction(TheStore, *SE, TLI); ++NumMemSet; return true; } @@ -579,7 +581,7 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, getAnalysis(), SI)) { Expander.clear(); // If we generated new code for the base pointer, clean up. - deleteIfDeadInstruction(StoreBasePtr, *SE); + deleteIfDeadInstruction(StoreBasePtr, *SE, TLI); return false; } @@ -594,8 +596,8 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, StoreSize, getAnalysis(), SI)) { Expander.clear(); // If we generated new code for the base pointer, clean up. - deleteIfDeadInstruction(LoadBasePtr, *SE); - deleteIfDeadInstruction(StoreBasePtr, *SE); + deleteIfDeadInstruction(LoadBasePtr, *SE, TLI); + deleteIfDeadInstruction(StoreBasePtr, *SE, TLI); return false; } @@ -628,7 +630,7 @@ processLoopStoreOfLoopLoad(StoreInst *SI, unsigned StoreSize, // Okay, the memset has been formed. Zap the original store and anything that // feeds into it. - deleteDeadInstruction(SI, *SE); + deleteDeadInstruction(SI, *SE, TLI); ++NumMemCpy; return true; } diff --git a/lib/Transforms/Scalar/LoopInstSimplify.cpp b/lib/Transforms/Scalar/LoopInstSimplify.cpp index 982400c..f5daa7b 100644 --- a/lib/Transforms/Scalar/LoopInstSimplify.cpp +++ b/lib/Transforms/Scalar/LoopInstSimplify.cpp @@ -120,7 +120,7 @@ bool LoopInstSimplify::runOnLoop(Loop *L, LPPassManager &LPM) { ++NumSimplified; } } - LocalChanged |= RecursivelyDeleteTriviallyDeadInstructions(I); + LocalChanged |= RecursivelyDeleteTriviallyDeadInstructions(I, TLI); if (IsSubloopHeader && !isa(I)) break; diff --git a/lib/Transforms/Utils/BasicBlockUtils.cpp b/lib/Transforms/Utils/BasicBlockUtils.cpp index 2679b93..75a7817 100644 --- a/lib/Transforms/Utils/BasicBlockUtils.cpp +++ b/lib/Transforms/Utils/BasicBlockUtils.cpp @@ -94,7 +94,7 @@ void llvm::FoldSingleEntryPHINodes(BasicBlock *BB, Pass *P) { /// is dead. Also recursively delete any operands that become dead as /// a result. This includes tracing the def-use list from the PHI to see if /// it is ultimately unused or if it reaches an unused cycle. -bool llvm::DeleteDeadPHIs(BasicBlock *BB) { +bool llvm::DeleteDeadPHIs(BasicBlock *BB, const TargetLibraryInfo *TLI) { // Recursively deleting a PHI may cause multiple PHIs to be deleted // or RAUW'd undef, so use an array of WeakVH for the PHIs to delete. SmallVector PHIs; @@ -105,7 +105,7 @@ bool llvm::DeleteDeadPHIs(BasicBlock *BB) { bool Changed = false; for (unsigned i = 0, e = PHIs.size(); i != e; ++i) if (PHINode *PN = dyn_cast_or_null(PHIs[i].operator Value*())) - Changed |= RecursivelyDeleteDeadPHINode(PN); + Changed |= RecursivelyDeleteDeadPHINode(PN, TLI); return Changed; } diff --git a/lib/Transforms/Utils/Local.cpp b/lib/Transforms/Utils/Local.cpp index bed7d72..0601433 100644 --- a/lib/Transforms/Utils/Local.cpp +++ b/lib/Transforms/Utils/Local.cpp @@ -52,7 +52,8 @@ using namespace llvm; /// Also calls RecursivelyDeleteTriviallyDeadInstructions() on any branch/switch /// conditions and indirectbr addresses this might make dead if /// DeleteDeadConditions is true. -bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions) { +bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions, + const TargetLibraryInfo *TLI) { TerminatorInst *T = BB->getTerminator(); IRBuilder<> Builder(T); @@ -96,7 +97,7 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions) { Value *Cond = BI->getCondition(); BI->eraseFromParent(); if (DeleteDeadConditions) - RecursivelyDeleteTriviallyDeadInstructions(Cond); + RecursivelyDeleteTriviallyDeadInstructions(Cond, TLI); return true; } return false; @@ -161,7 +162,7 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions) { Value *Cond = SI->getCondition(); SI->eraseFromParent(); if (DeleteDeadConditions) - RecursivelyDeleteTriviallyDeadInstructions(Cond); + RecursivelyDeleteTriviallyDeadInstructions(Cond, TLI); return true; } @@ -205,7 +206,7 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions) { Value *Address = IBI->getAddress(); IBI->eraseFromParent(); if (DeleteDeadConditions) - RecursivelyDeleteTriviallyDeadInstructions(Address); + RecursivelyDeleteTriviallyDeadInstructions(Address, TLI); // If we didn't find our destination in the IBI successor list, then we // have undefined behavior. Replace the unconditional branch with an @@ -230,7 +231,8 @@ bool llvm::ConstantFoldTerminator(BasicBlock *BB, bool DeleteDeadConditions) { /// isInstructionTriviallyDead - Return true if the result produced by the /// instruction is not used, and the instruction has no side effects. /// -bool llvm::isInstructionTriviallyDead(Instruction *I) { +bool llvm::isInstructionTriviallyDead(Instruction *I, + const TargetLibraryInfo *TLI) { if (!I->use_empty() || isa(I)) return false; // We don't want the landingpad instruction removed by anything this general. @@ -265,9 +267,9 @@ bool llvm::isInstructionTriviallyDead(Instruction *I) { return isa(II->getArgOperand(1)); } - if (isAllocLikeFn(I)) return true; + if (isAllocLikeFn(I, TLI)) return true; - if (CallInst *CI = isFreeCall(I)) + if (CallInst *CI = isFreeCall(I, TLI)) if (Constant *C = dyn_cast(CI->getArgOperand(0))) return C->isNullValue() || isa(C); @@ -278,9 +280,11 @@ bool llvm::isInstructionTriviallyDead(Instruction *I) { /// trivially dead instruction, delete it. If that makes any of its operands /// trivially dead, delete them too, recursively. Return true if any /// instructions were deleted. -bool llvm::RecursivelyDeleteTriviallyDeadInstructions(Value *V) { +bool +llvm::RecursivelyDeleteTriviallyDeadInstructions(Value *V, + const TargetLibraryInfo *TLI) { Instruction *I = dyn_cast(V); - if (!I || !I->use_empty() || !isInstructionTriviallyDead(I)) + if (!I || !I->use_empty() || !isInstructionTriviallyDead(I, TLI)) return false; SmallVector DeadInsts; @@ -301,7 +305,7 @@ bool llvm::RecursivelyDeleteTriviallyDeadInstructions(Value *V) { // operand, and if it is 'trivially' dead, delete it in a future loop // iteration. if (Instruction *OpI = dyn_cast(OpV)) - if (isInstructionTriviallyDead(OpI)) + if (isInstructionTriviallyDead(OpI, TLI)) DeadInsts.push_back(OpI); } @@ -334,19 +338,20 @@ static bool areAllUsesEqual(Instruction *I) { /// either forms a cycle or is terminated by a trivially dead instruction, /// delete it. If that makes any of its operands trivially dead, delete them /// too, recursively. Return true if a change was made. -bool llvm::RecursivelyDeleteDeadPHINode(PHINode *PN) { +bool llvm::RecursivelyDeleteDeadPHINode(PHINode *PN, + const TargetLibraryInfo *TLI) { SmallPtrSet Visited; for (Instruction *I = PN; areAllUsesEqual(I) && !I->mayHaveSideEffects(); I = cast(*I->use_begin())) { if (I->use_empty()) - return RecursivelyDeleteTriviallyDeadInstructions(I); + return RecursivelyDeleteTriviallyDeadInstructions(I, TLI); // If we find an instruction more than once, we're on a cycle that // won't prove fruitful. if (!Visited.insert(I)) { // Break the cycle and delete the instruction and its operands. I->replaceAllUsesWith(UndefValue::get(I->getType())); - (void)RecursivelyDeleteTriviallyDeadInstructions(I); + (void)RecursivelyDeleteTriviallyDeadInstructions(I, TLI); return true; } } @@ -358,7 +363,8 @@ bool llvm::RecursivelyDeleteDeadPHINode(PHINode *PN) { /// /// This returns true if it changed the code, note that it can delete /// instructions in other blocks as well in this block. -bool llvm::SimplifyInstructionsInBlock(BasicBlock *BB, const TargetData *TD) { +bool llvm::SimplifyInstructionsInBlock(BasicBlock *BB, const TargetData *TD, + const TargetLibraryInfo *TLI) { bool MadeChange = false; #ifndef NDEBUG @@ -381,7 +387,7 @@ bool llvm::SimplifyInstructionsInBlock(BasicBlock *BB, const TargetData *TD) { continue; } - MadeChange |= RecursivelyDeleteTriviallyDeadInstructions(Inst); + MadeChange |= RecursivelyDeleteTriviallyDeadInstructions(Inst, TLI); if (BIHandle != BI) BI = BB->begin(); } diff --git a/lib/Transforms/Utils/SimplifyInstructions.cpp b/lib/Transforms/Utils/SimplifyInstructions.cpp index 81eb9e0..528e6a1 100644 --- a/lib/Transforms/Utils/SimplifyInstructions.cpp +++ b/lib/Transforms/Utils/SimplifyInstructions.cpp @@ -72,7 +72,7 @@ namespace { ++NumSimplified; Changed = true; } - Changed |= RecursivelyDeleteTriviallyDeadInstructions(I); + Changed |= RecursivelyDeleteTriviallyDeadInstructions(I, TLI); } // Place the list of instructions to simplify on the next loop iteration diff --git a/lib/Transforms/Vectorize/BBVectorize.cpp b/lib/Transforms/Vectorize/BBVectorize.cpp index 62d23cb..c09dcd2 100644 --- a/lib/Transforms/Vectorize/BBVectorize.cpp +++ b/lib/Transforms/Vectorize/BBVectorize.cpp @@ -601,7 +601,7 @@ namespace { // It is important to cleanup here so that future iterations of this // function have less work to do. - (void) SimplifyInstructionsInBlock(&BB, TD); + (void) SimplifyInstructionsInBlock(&BB, TD, AA->getTargetLibraryInfo()); return true; } diff --git a/test/Transforms/GVN/malloc-load-removal.ll b/test/Transforms/GVN/malloc-load-removal.ll new file mode 100644 index 0000000..66b6929 --- /dev/null +++ b/test/Transforms/GVN/malloc-load-removal.ll @@ -0,0 +1,31 @@ +; RUN: opt -S -basicaa -gvn < %s | FileCheck %s +; RUN: opt -S -basicaa -gvn -disable-simplify-libcalls < %s | FileCheck %s -check-prefix=CHECK_NO_LIBCALLS +; PR13694 + +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx10.8.0" + +declare i8* @malloc(i64) nounwind + +define noalias i8* @test() nounwind uwtable ssp { +entry: + %call = tail call i8* @malloc(i64 100) nounwind + %0 = load i8* %call, align 1 + %tobool = icmp eq i8 %0, 0 + br i1 %tobool, label %if.end, label %if.then + +if.then: ; preds = %entry + store i8 0, i8* %call, align 1 + br label %if.end + +if.end: ; preds = %if.then, %entry + ret i8* %call + +; CHECK: @test +; CHECK-NOT: load +; CHECK-NOT: icmp + +; CHECK_NO_LIBCALLS: @test +; CHECK_NO_LIBCALLS: load +; CHECK_NO_LIBCALLS: icmp +} -- cgit v1.1